The document summarizes the IBM z Systems z13 mainframe update. Key points include the status of IBM servers and trends in digital disruption driving increased mainframe requirements. The z13 launch is highlighted as enabling lower costs through improvements like simultaneous multithreading and large memory capabilities. Mainframes are described as the platform for the future, processing a growing number of mobile transactions worldwide and supporting a large portion of critical applications.
1. 12/10/2015
1
Large Systems Update (LSU) 2015
z Systems z13 update
Status, Trends and Directions
KMD/JN Data Århus, December 8
The Modern Mainframe Redefining Digital Business,
…at the brink of the Cognitive Era.
Version 5.9 December 2015
January 14, 2015 Announcement
Henrik Thorsen, IBM Technical Director
Nordic z Systems Platform Leader
With credit for certain charts to fellow IBMers
Trademarks
Notes:
Performance is in Internal Throughput Rate (ITR) ratio based on measurements and projections using standard IBM benchmarks in a controlled environment. The actual throughput that any user will experience will vary
depending upon considerations such as the amount of multiprogramming in the user's job stream, the I/O configuration, the storage configuration, and the workload processed. Therefore, no assurance can be given that an
individual user will achieve throughput improvements equivalent to the performance ratios stated here.
IBM hardware products are manufactured from new parts, or new and serviceable used parts. Regardless, our warranty terms apply.
All customer examples cited or described in this presentation are presented as illustrations of the manner in which some customers have used IBM products and the results they may have achieved. Actual environmental
costs and performance characteristics will vary depending on individual customer configurations and conditions.
This publication was produced in the United States. IBM may not offer the products, services or features discussedin this document in other countries, and the information may be subject to change without notice. Consult
your local IBM business contact for information on the product or services available in your area.
All statements regarding IBM's future direction and intent are subject to change or withdrawal without notice, and represent goals and objectives only.
Information about non-IBM products is obtained from the manufacturers of those products or their published announcements. IBM has not tested those products and cannot confirm the performance, compatibility, or any
other claims related to non-IBM products. Questions on the capabilities of non-IBM products should be addressed to the suppliers of those products.
Prices subject to change without notice. Contact your IBM representative or Business Partner for the most current pricing in your geography.
This informationprovides only general descriptions of the types and portions of workloads that are eligible for execution on Specialty Engines (e.g, zIIPs, zAAPs, and IFLs) ("SEs"). IBM authorizes customers to use IBM SE
only to execute the processing of Eligible Workloads of specific Programs expressly authorized by IBM as specified in the “Authorized Use Table for IBM Machines” provided at
www.ibm.com/systems/support/machine_warranties/machine_code/aut.html (“AUT”). No other workload processing is authorized for execution on an SE. IBM offers SE at a lower price than General
Processors/CentralProcessors because customers are authorized to use SEs only to process certain types and/or amounts of workloads as specified by IBM in the AUT.
* Registered trademarksof IBM Corporation
* Other product and service names might be trademarksof IBM or other companies.
Adobe, the Adobe logo, PostScript, and the PostScript logo are either registered trademarks or trademarks of Adobe Systems Incorporated in the United
States, and/or other countries.
IT Infrastructure Library is a registered trademark of the Central Computer and Telecommunications Agency which is now part of the Office of
Government Commerce.
Intel, Intel logo, Intel Inside, Intel Inside logo, Intel Centrino, Intel Centrino logo, Celeron, Intel Xeon, Intel SpeedStep, Itanium, and Pentium are
trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States and other countries.
Linux is a registered trademark of Linus Torvalds in the United States, other countries, or both.
Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft Corporation in the United States, other countries, or both.
Windows Server and the Windows logo are trademarks of the Microsoft group of countries.
ITIL is a registered trademark, and a registered community trademark of the Office of Government Commerce, and is registered in the U.S. Patent and
Trademark Office.
UNIX is a registered trademark of The Open Group in the United States and other countries.
Java and all Java based trademarks and logos are trademarks or registered trademarks of Oracle and/or its affiliates.
Cell Broadband Engine is a trademark of Sony Computer Entertainment, Inc. in the United States, other countries, or both and is used under license
therefrom.
Linear Tape-Open, LTO, the LTO Logo, Ultrium, and the Ultrium logo are trademarks of HP, IBM Corp. and Quantum in the U.S. and other countries.
The following are trademarks of the International Business Machines Corporation in the United States and/or other countries.
The following are trademarks or registered trademarks of other companies.
2. 12/10/2015
2
• The following people “contributed” to this presentation:
• Riaz Ahmad, Matthias Bangert, Uno Bengtson, Mario Bezzi, Maria K Boisen, Nick Clayton,
• Donna Dillenberg, Martin Dvorsky, Michael Eggloff, Harv Emery, Cindy Grossman, Ray Jones,
• Frank Kyne (Watson & Walker), Parwez Hamid, Gerard Laumay, Helene Lyon, Silvia Melitta,
• Marianne Menå Heltborg, Frank Packheiser, Ewerson Palacio, Alain Poquillon,
• Per Rosenquist, Jørgen Riis Andersen, Jeff Seidell, Harri Stranden, Peter Sommer
• Christopher Spaight, Svenn-Aage Sønderskov (BEC), Henrik Thorsen, Robert Vaupel,
• Dan Wardman, Charles Webb
• ... and many many more
...Irv Gordon (”5 million km Volvo man”),
....Per Groth (”2 Volvo book writer”)
Acknowledgements Thanks!
3
Agenda
4
IT world transforming fast, creating extra MF requirements
• Consumer, Economic and IT MF perspective in ww, European, Nordic context
• Digital disruption, CAMSS, IoT, Cognitive Era
IBM servers holistic view – survival of the fittest
• IBM divesting 86-based servers, emphasizing z Systems and POWER brands
• What about Moore’s law? Silicon Semiconductor update
• Innovation much more than semiconductors and HW
IBM z Systems z13 enabling lower cost:
• Simultaneous Multithreading (SMT)
• Single Instruction Multiple Data (SIMD)
• Large Memory and Memory Affinity
• Significant I/O, Security and other enhancements
IBM z13 and Real Time Transaction/Predictive Analytics
IBM client value of HW/SW currency
IBM z13 Status:
• Obervations, tips and tricks, tools, videos
• workshops – ITSO, LinuxOne, GSE,...
3. 12/10/2015
3
5
2014 Key IBM Systems Server Milestones
6
2015 Key IBM Systems Server Milestones
Systems HW Ongoing Transformation – 2015 Milestones
January
z13 launch
July
IBM launch 7 nm chip,
world’s smallest:
GlobalFoundries and
Samsung
March
3$B R&D
investment in
IoT technology
LinuxONE
Emperor,
world’s most
advanced Linux
system
Welcome to the
cognitive era
IBM CEO Rometty
describes new era in
technology
August -YE
October
IBM POWER8
4. 12/10/2015
4
7
The 3rd generation of computing platform, the 3rd phase
of the Internet, and the explosion of information are colliding to
form a perfect storm of disruption and transformation
2020-2
Amount of Data
Collected and
Stored
Adapted from HorizonWatch: Top Technology Trends To Watch In 2013 Source: Bill Chamberlin
1964 1981 1994 2003 2008 2012
Continuum of
Computing
Platforms
Mainframe
Client Server / PC
Mobile Devices
Internet Web 2.0
Phases of
the Internet Web 3.0 (Cloud, Analytics, Mobile, Social)
IoT
8
The 3rd generation of computing platform, the 3rd phase
of the Internet, and the explosion of information are colliding to
form a perfect storm of disruption and transformation
The Internet of Things – infographic The Connectivist based on Cisco data -2014
Source: Slide Share:
Ammar Sabzwari
5. 12/10/2015
5
Eras of Computing
9
By 2018, 1/3rd of leaders in each industry will
be disrupted by 3rd platform competitors.
The 3rd Platform allows
businesses to:
• Create greater operational
efficiencies
• Build deeper relationships with their
customers
• Create new revenue streams
based on technology-enabled
products and services
The 3rd Platform is a Business Platform
10
6. 12/10/2015
6
Digital Disruption is Real
11
75% of S&P 500 companies may be replaced by 2027!
> 60 years
< 20 years
1st Platform 2nd Platform 3rd Platform
Digital Disruption is Real
12
Everything in
This 1991
Radio Shack
Ad Has Been
Replaced w/
Apps Running
on a
Smartphone
8. 12/10/2015
8
Eagle studies validate that adding workload to
Mainframes reduces labor cost per unit of work
15
Mobile, Mobile, Mobile!
Transactions generate business, MF load,
and they are growing
16
Mainframes in an Worldwide context:
• Recent past - 10 years ago
• Less than 1 transaction a day
• Today – right now
• Up to 37 transactions a day
• Future – 3 years and in 10 years
• 2017 50 trans/day
• 2025 1.6 Trillion+ trans/day from 10
Billion devices
=1.600.000.000.000 transactions
30.000.000.000 MF business
transaction today
50% Compound annual growth rate
Sources:
Juniper, Gartner, Wall Street
Journal Japan, IBM PoV
9. 12/10/2015
9
The Platform for the Future
IDC says 60% critical apps run on IBM MF today
17
Mainframes in an European context:
• 10.000 mobile transactions happened when
you read the heading!
• Drive 12% of Europe's economy
• Generate more than 110,000 jobs
• Creates an ecosystem of 1000 partner
organizations
• €5.6 billion ecosystem revenue
• Open architectures and Linux are sizzling
18
20M+ 100B 6x and 3x
apps in the
world today
apps was
downloaded in 2014
Google and Apple respectively have
released more major Android and iOS
versions than Microsoft has released
major Windows PC versions
Build and Connect
System z mobile web, hybrid,
and native app development
System z data, service and
application integration
Lifecycle management
Building and connecting System z data to mobile devices
to provide a better, more-secure customer experience
Building and connecting apps to z Systems
10. 12/10/2015
10
The Platform for the Future
IDC says 60% critical apps run on IBM MF today
19
Mainframes in a Nordic context:
• ~ 90 Mainframes
• Surrounded by ~ 900.000+ other servers
• ~ 600.000 MIPs
• ~ 50% on latest z13 technology (~ 50% YE)
• ~ 2% of Worldwide MF server capacity
• Growth rate MIPs (5 YR CAGR) ~ 15%
• More for specialty engines
• Linux on z MIPs (5 YR) ~ 45%!
• ~ 1.000.000.000 MF business transactions/day
• ~ 50.000 MF business transactions/sec in peak
• MFs are vital to Nordic cooperations, our society, you,
IBM and myself
• Latest version MF z13 announced January 14, 2015
...continues to carry the torch
The Platform for the Future
IDC says 60% critical apps run on IBM MF today
20
Mainframes in a Nordic context:
• ~ 90 Mainframes
• Surrounded by ~ 900.000+ other servers
• ~ 600.000 MIPs
• ~ 30% on latest z13 technology (~ 50% YE)
• ~ 2% of Worldwide MF server capacity
• Growth rate MIPs (5 YR CAGR) ~ 15%
• More for specialty engines
• Linux on z MIPs (5 YR) ~ 45%!
• ~ 1-2.000.000.000 MF business transactions/day
• ~ 50.000 MF business transactions/sec in peak
• MFs are vital to Nordic cooperations, our society, you,
IBM and myself
• Latest version MF z13 announced January 14, 2015
Mainframes in a Worldwide context:
11. 12/10/2015
11
The Platform for the Future
IDC says 60% critical apps run on IBM MF today
21
Mainframes in a Nordic context:
• ~ 90 Mainframes
• Surrounded by ~ 900.000+ other servers
• ~ 600.000 MIPs
• ~ 30% on latest z13 technology (~ 50% YE)
• ~ 2% of Worldwide MF server capacity
• Growth rate MIPs (5 YR CAGR) ~ 15%
• More for specialty engines
• Linux on z MIPs (5 YR) ~ 45%!
• ~ 1-2.000.000.000 MF business transactions/day
• ~ 50.000 MF business transactions/sec in peak
• MFs are vital to Nordic cooperations, our society, you,
IBM and myself
• Latest version MF z13 announced January 14, 2015
Market Forces
2014-2018
22
2018 Devices:
• 2x growth
• 40 billion
• 5.0/person
2018 Mobile Users:
• 50% growth
• 3.8 billion
• 50% penetration
… yesterday’s infrastructure won’t cut it…
2018 Datacenter Cores:
• 2x growth
• 77 billion
• 10/person
2018 Data:
• 3x growth
• 24 zettabytes
• 1.5 TBs/person/day
12. 12/10/2015
12
IBM z Systems z13 Platform Positioning
• The world’s premier data and
transaction engine enabled for
the mobile generation
• The integrated transaction and
analytics system for right-time
insights at the point of impact
• The world’s most efficient and
trusted cloud system that
transforms the economics of IT
Transaction Processing
Data Serving
Mixed Workloads
Operational Efficiency
Trusted and Secure Computing
Reliable, Available, Resilient
Virtually Limitless Scale
2323
Enable superior Cloud services at up to
32% lower cost than x86 Cloud and up to
60% less than Public Cloud over three years
Deliver up to 36% better response time,
up to 61% better throughput, and 17 to
37% lower cost per mobile transaction
Deliver insights up to 17x faster and
with 13x better price performance than
closest competitor
Accelerate speed of encryption
up to 2X over the zEC12 to help
protect the privacy of data throughout
its life cycle
Cloud
Analytics
Mobile
Security
The all new IBM z13:
Excel in Digital Business Against Competition
13. 12/10/2015
13
Innovation Drives Performance
0%
20%
40%
60%
80%
100%
180 nm 130 nm 90 nm 65 nm 45 nm 32 nm 22 nm
Gain by Technology Scaling Gain by Innovation
Relative %
of Improvement
25
• Future growth also comes from dimensions other than hardware thread speed alone:
• Software efficiency – extract full performance value from System z hardware
– Compiler technology, dynamic optimization, exploit new architectural facilities
• Core density – multi-core processor chips
– More efficient use of space, less complex packaging
• Cache density – leveraging leadership eDRAM technology
– Larger caches closer to more cores
Combined result: continued growth in system capacity each generation
• Sysplex scale – more systems sharing data and workloads
• Workload optimization via specialized capabilities
Compute Capability Trends
26
14. 12/10/2015
14
Java
Semiconductor
Technology
Microprocessor
Design
Systems
Design
Virtualization
& Operating
Systems
Compilers &
Java Virtual
Machine
Optimized
Middleware
IBM z Systems z13
27
IBM $3B R&D investment
in chip technologies over next 5 years
IBM is tackling chip challenges by launching
two broad research and development
programs:
1. 7nm and beyond silicon technology will
address serious physical challenges
that are threatening current
semiconductor scaling techniques
2. Alternative technologies for post-silicon
era chips under development:
• Silicon Photonics/Nanowire
• Silicium Germanium...
• (2012 9nm transistor)
• (2015 7nm experimental chip capable
of holding up to 20 Billion transistors)
• Carbon Nanotubes
• Graphene
• Neurosynaptic Computing
• Quantum Computing
28
15. 12/10/2015
15
IBM $3B R&D investment
in chip technologies over next 5 years
IBM is tackling chip challenges by launching
two broad research and development
programs:
1. 7NM and beyond silicon technology
will address serious physical
challenges that are threatening current
semiconductor scaling techniques
2. Alternative technologies for post-silicon
era chips under development:
• Silicon Photonics/Nanowire
• Silicium Germanium...
• Carbon Nanotubes
• (2012 9NM transistor)
• (2015 7NM chip))
• Graphene
• Neurosynaptic Computing
• Quantum Computing - Qubits
29
IBM is tackling chip challenges by launching
two broad research and development
programs:
1. 7NM and beyond silicon technology
will address serious physical
challenges that are threatening current
semiconductor scaling techniques
2. Alternative technologies for post-silicon
era chips under development:
• Silicon Photonics/Nanowire
• Silicium Germanium...
• Carbon Nanotubes
• (2012 9NM transistor)
• (2015 7NM chip))
• Graphene
• Neurosynaptic Computing
• Quantum Computing
IBM $3B R&D investment
in chip technologies over next 5 years
30
5 year development
effort, 1B$ and 500
patents/year involved
in launch of z13
16. 12/10/2015
16
31
z13 - Refined z Systems Architecture
• Performance: How fast a given piece of work can be done?
• Throughput: How much work can be done in a given amount of time?
• Performance and throughput are the combination of architecture, design innovation and
processor frequency (cycle time)
• z13 improvements delivered through:
– Design innovations such as increased cache sizes, Out of Order instruction processing,
restructured pipelines, increased n-way design, etc.
– Increasing Uni-processor performance with a reduction in cycle time when compared to zEC12
– Increasing the scale in the system N-way
– Introducing Simultaneous Multi Threading (SMT)
– Reintroducing SIMD vector processing
– Extending memory from 3TB to 10TB
Frequency Scaling
x86, POWER, z Systems
x86 and POWER already transitioned to a throughput-centric model
Frequency peaked for x86 in 2005, POWER in 2008
In System z we have held this off with unique cooling, packaging, technology, and design
solutions
Frequency(MHz)
0
1000
2000
3000
4000
5000
6000
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
x86
Power
System z
17. 12/10/2015
17
CPU core speed vs Computer Performance
Why the overall CPU frequency approach is changing ?
̶ Consistent frequency growth in the past decade
• from hundreds of MHz to GHz
̶ CPU frequency has reduced in the past couple of years
Designing chips for better performance
̶ Limits are imposed by physics, technology or economics
̶ Controls the rate of improvements in different dimensions
̶ Different processor architectures have different issues with overclocking
Physical limitations
̶ Speed of light limits how fast signals travel form one end to the other on a chip
̶ Power and heat dissipation
̶ Cooling
̶ How many memory elements (caches) can be within a given latency from the CPU
Physical limitations force the designers to make trade-offs
̶ “Shrinking” a processor chip
• pro: Faster due to the shorter distances
• con: Reduced area for dissipation
Power dissipation increases as the chip speeds up
̶ Raising the processor voltages would make transistors to switch quicker
• pro: Frequency could then be increased
• con: current also increases creating more heat
Sounds easy.. but… it causes serious problems with heat
Emerging technologies allow frequency variation according to processing needs
CPU Clock speed versus Computer Performance…
GHz is not the only dimension that matters
̶ System z focus is on balanced system design across many factors:
• Frequency, pipeline, efficiency, energy efficiency, cache/memory design and I/O design
• Greater logic density, power density, wire-ability. All permits more cores per chip, larger
cache, additional execution units/circuits, addition of SMT and SIMD on each core.
System performance is not linear with frequency
̶ Need to use LSPR and System z capacity planning tools for real client / workload sizing
System z leverages technologies to get the most out of chips design
̶ Low latency pipelines
̶ Dense packaging with proper cooling which yields more power-efficient operation
̶ Consistent performance at high utilization
The IBM z13 Server is a significant change from zEC12
̶ Processor speed measured in instructions per second (for a given workload) has increased as
compared to the zEC12.
• Wider pipeline (up to six per cycle)
• Enhanced branch prediction
• Optimized resolution of dependencies between instructions.
18. 12/10/2015
18
z Systems - Processor Roadmap
z196
9/2010
zEC12
8/2012
z10
2/2008
z13
1/2015
Leadership Single Thread,
Enhanced Throughput
Improved out-of-order
Transactional Memory
Dynamic Optimization
2 GB page support
Step Function in System
Capacity
Top Tier Single Thread
Performance System Capacity
Accelerator Integration
Out of Order Execution
Water Cooling
PCIe I/O Fabric
RAIM
Enhanced Energy Management
Leadership System Capacity and
Performance
Modularity & Scalability
Dynamic SMT
Supports two instruction threads
SIMD
PCIe attached accelerators (XML)
Business Analytics Optimized
Workload Consolidation and
Integration Engine for CPU
Intensive Workloads
Decimal FP
Infiniband
64-CP Image
Large Pages
Shared Memory
8 double-wide cores per CP chip
2X Instruction pipe width
– Improves IPC for all modes
– Symmetry simplifies dispatch/issue rules
– Required for effective SMT
Added FXU and BFU execution units
– 4 FXUs (Fixed point units)
– 2 BFUs (Binary floating-point units)
– DFUs (Decimal floating-point units)
– 2 new SIMD units
SIMD unit plus additional registers
Pipe depth re-optimized for
power/performance
– Product frequency reduced (5.0 GHz)
– Processor performance increased
SMT2 support
– Wide, symmetric pipeline
– Full architected state per thread
– SMT-adjusted CPU usage metering
z13 Processor Overview
Enhancements at a glance
19. 12/10/2015
19
IFB
ICM
LSU
L2D (eDRAM)
RU
IDU
ISU
XU TU
VFU
FXU
IFB
ICM
LSU
ISU
IDU
FXU
RU
L2D
L2I
XU
PC
VFU
COP
Where is the SIMD Unit?
Fixed Point Unit
• Counters (Loops)
• Adresses
Vector & Float Unit
• Fixed Point Decimal (BCD)
• Floating Point HEX
• Floating Point BINARY
• Floating Point DECIMAL
• Vector string processing
• Binary vector integer
z13 VFU (Vector and Floating-Point Units)
• Enhancements
– Two execution pipelines & 4x FPU / VFU registers
– SIMD & string (new engine)
• String processing & vector execution of binary integer
– DFX (decimal FXU)
• new engine optimized for BCD add/sub/compare
• zEC12: these ops executed on DFU
– BFU: Binary & Hex floating-point unit
• new: Vector support for Binary 64b float
– Leading edge Divide/sqrt engine for binary & hex FP
• zEC12: divide & square-root executed in the BFU
• Shorter latency and higher throughput than zEC12
– DFU: decimal floating-point unit
• 3rd generation, allowing for higher throughput
• Executes BCD mul, div, convert, shift
– All Units are implemented twice – so in the best case this doubles
througput compared to zEC12
BFU
BFU
Divide
Divide
DFU
DFU
DFX
SIMD
&
String
DFX
SIMD
&
String
RegFiles
RegFiles
zEC12 FPUs
Silvia M Mueller, Jan/2015
20. 12/10/2015
20
New Memory Controller
Crypto Express5S
FICON Express16S
1U Support Element
Standalone zBX Node Hybrid Computing
2.7M lines of firmware changed
Radiator Design improvements
Expanded operating environment (Rear
Doors)
22nm Processor with SIMD, SMT
Integrated I/O with PCIe Direct Attach
Single Chip Modules
Drawer-Based CPC Design
Cable-Based SMP Fabric
Oscillator Backplane
Flexible Service Processor (FSP2)
Integrated Sparing
On-chip power/thermal monitor /
control
z13 System Design Changes
New Memory Controller
Crypto Express5S
FICON Express16S
1U Support Element
Standalone zBX Node Hybrid Computing
2.7M lines of firmware changed
Radiator Design improvements
Expanded operating environment (Rear
Doors)
22nm Processor with SIMD, SMT
Integrated I/O with PCIe Direct
Attach
Single Chip Modules
Drawer-Based CPC Design
Cable-Based SMP Fabric
Oscillator Backplane
Flexible Service Processor (FSP2)
Integrated Sparing
On-chip power/thermal monitor /
control
z13 System Design Changes
21. 12/10/2015
21
Physical node: (Two per drawer)
Chips
– Three PU chips (with up to 8 active cores each)
– One SC chip (480 MB L4 cache)
RAIM Memory
– Three Memory Controllers: One per CP Chip
– Five DDR3 DIMM slots per Controller: 15 total per logical node
– Populated DIMM slots: 20 or 25 per drawer
SC and CP Chip Interconnects
– X-bus: SC and CPs to each other (intra node)
– S-bus: SC to SC chip in the (intra drawer)
– A-bus: SC to SC chips in the remote drawers (intra box)
Mem
PSI
Mem
GX Bus
2x PCIe
GX Bus
2x PCIePUPU
SCSC
MemMem DIMMs
GX Bus
2x PCIe
Node 1
GX Bus
2x PCIe
GX Bus
2x PCIe
GX Bus
2x PCIe
Fully Populated Drawer
MemMem
A-Bus
S-Bus
X-Bus
Node 0
X-Bus
SCSC
A-Bus
To other
drawers
To other
drawers
z13 Drawer Structure and Interconnect
Drawer Based SMP Topology
Node 1 Node 0
Node 1 Node 0
Node 1 Node 0
Node 1 Node 0
Drawer
Drawer
Drawer
Drawer
PU
PU
PU PU
PU
z13 versus zEC12 HW Comparison
… and z196
zEC12 (z196)
– CPU
• 5.5 GHz (1514 PCI)
• Enhanced Out-Of-Order
– Caches
• L1 private 64k i, 96k d (Same)
• L2 private 1 MB i + 1 MB d (1.5 MB)
• L3 shared 48 MB / chip (24 MB)
• L4 shared 384 MB / book (192 MB)
z13
– CPU
• 5.0 GHz (1695 PCI)
• Major pipeline enhancements
– Caches
• L1 private 96k i, 128k d
• L2 private 2 MB i + 2 MB d
• L3 shared 64 MB / chip
• L4 shared 480 MB / node
- plus 224 MB NIC
...
Memory
L4 Cache
L2
CPU1
L1
L3 Cache
L2
CPU6
L1... L2
CPU1
L1
L3 Cache
L2
CPU6
L1...
...
Memory
L4 Cache
L2
PU1
L1
L3 Cache
... L2
PU8
L1
L2
PU1
L1
L3 Cache
...L2
PU8
L1
...
Memory
L4 Cache
L2
PU1
L1
L3 Cache
... L2
PU8
L1
L2
PU1
L1
L3 Cache
...L2
PU8
L1
Single Book View
Single Drawer View
22. 12/10/2015
22
z13 Architecture Extensions
Core, SIMD, SMT-2
• Core micro-architecture radically altered to increase parallelism
– New branch prediction and instruction fetch front end
• Supports SMT-2 and improves branch prediction throughput for all modes
– Wider instruction decode, dispatch and completion bandwidth:
• Up to six instructions per cycle compared to three on zEC12
– Up to ten instructions issued for execution per cycle compared to seven on zEC12
• 2 branch, 4 FXU, 2 LSU, 2 BFU/DFU/SIMD
• Single Instruction Multiple Data (SIMD) instruction set and execution:
Business Analytics Vector Processing
– 139 new instructions operate on 32 new 128-bit registers
– String, vector integer and vector floating point operations
• Two 64-bit, four 32-bit, eight 16-bit, or 8-bit operands per register
• Two-way simultaneous multithreaded (SMT-2) operation
– Up to two active execution threads per core
• Dynamically share caches, TLBs, and execution resources.
• Significant boost in core and chip throughput on top of core performance increase
– SMT-2 operation supported for IFLs and zIIPs
– Hardware monitoring support for chargeback and capacity planning
43
Simultaneous Multi-threading (SMT)
• Simultaneous Multi-threading (SMT) technology
• Multiple programs (software threads) run on same processor core
• More efficient use of core hardware
• Active threads share core resources
• In space: data and instruction caches, TLBs, branch history tables, etc.
• In time: pipeline slots, execution units, address translator, etc.
• Increases overall throughput per core when SMT active
• Amount of increase varies widely with workload – typically 1.2-1.6X
• Each thread runs more slowly than on a single-thread core
A
/
B
A
B
B
Load/Store (L1 Cache)
A A B
Execution Units (FXU/FPU)
instructions
A B A A
B A
A
B
Shared Cache
A B A A
B A
BA
Cache
Thread-A
Thread-B
Use of Pipeline Stages in SMT-2
Both threads
Stage idle
44
23. 12/10/2015
23
Simultaneous Multi Threading (SMT2)
• Simultaneous multithreading allows instructions
from more than one thread to execute in any given
pipeline stage at a time
• SMT helps address memory latency, resulting in
overall throughput gains
• It can increase processing efficiency, and
throughput
• Currently available on IFLs (Linux and z/VM) and
zIIPs (z/OS)
• The number of concurrent threads is limited to two
and can be turned on or off by an operator
command and also set up through parmlib for z/OS
45
Which approach is designed
for the highest volume of
traffic? Which road is faster?
Illustrative numbers only
Note: SMT is designed to deliver better overall throughput for many workloads.
Performance in some cases may be superior using single threading
Why
only
45?
SMT dependency on Job runtime
• SMT increases the efficiency of a core. Instead of lets say running 100 instructions per
TIME in a single thread, SMT allows running more than 100 instructions in 2 threads in the
same TIME.
• However, sharing one ressource amoung 2 threads is „not for free“. There is a „cost“
associated.
• SMT Effectivness = (SMT Throughput / Single Thread Throughput).
– Lets assume we can run 125 instructions in the same timeintervall with SMT switched on we
would end up with SMT Eff. = 125 / 100
• The „cost“ is, that both threads run slower than a single thread.
• The elongation is calculated as: Elongation % = (2 / SMT Eff.) - 1
– In our example the single threads would run 60% slower than before.
• Lets assume your client has a system zEC12 with 10 ZIIPs.
– Because of z13‘s single thread performance improvement you only need 9 ZIIPs now
– With SMT switched on and 25% efficiency improvement you end up with 7 ZIIPs.
– In that case all workload on your ZIIPs will run 60% slower than without SMT and 46% slower
than before on zEC12.
24. 12/10/2015
24
Remarks on SMT
• For z/OS
– You enable SMT in LOAD xx at IPL
– Once done, enable/disable SMT per LPAR by dynamically activating different
OPT files.
• For native Linux
– There is no support (yet) in Redhat or Suse Linux for SMT.
– This is expected to be available in Q3 2015.
• For z/VM
– No dynamic switching of SMT modes
– Linux running under z/VM can exploit SMT
• Throughput improvements dependent on the workload with SMT:
– SAP runs very nicely with SMT switched on (large SMT throughput benefit)
– Imagine you are running OLTP transactions and time critical Batch Jobs on a
single core with SMT switched on – may be less beneficial (Bad Cache lines)
Enterprise Level Multithreading
Expands capacity without compromising predictability
• z13 delivers significant boost in IFL and zIIP performance (throughput and
capacity) via simultaneous multithreading (SMT)
• Extends per-processor capacity growth beyond single-thread performance
• z13 will support 2 threads per core
• Design will preserve unique System z values and attributes
• Predictable core capacity with precise utilization measurement
– Supports chargeback, billing, and capacity planning
• SMT enablement independently controlled by LPAR
• Operating systems must be explicitly enabled for SMT
• Operating system may opt to run in single-thread mode
• SMT control coordinated across hardware, hypervisors, software
• Operating system running in LPAR controls use of threads in each core
– Transparent to the application
• Hypervisors can leverage SMT for image consolidation
• Functionally transparent to middleware and applications
• Each hardware thread has full z/Architecture processor function
• No changes required to run in SMT partition
48
25. 12/10/2015
25
49
z13 and Simultaneous Multi Threading
• Simultaneous Multi Threading (SMT2) is available on IFLs and zIIPs only (currently)
– 10% to 35%+ throughput improvement on top of Uni-processor performance gains
• IFLs: 10% to 32%
• zIIPs:10% to 40%
– Will affect throughput (amount of work that can be done) versus performance (speeding up the IFL / zIIP)
• Be sensitive to workloads that need fastest possible single thread performance
– Individual work units may run slower due to each thread running slower on an SMT enabled core
• Example: If a core can do 1000 MIPS without SMT (one thread) and with SMT turned on runs 2 threads giving 40%
more throughput, then the core will deliver 1400 MIPS with SMT. Since the 1400 MIPS is the sum of the two threads,
each thread will be running at 1400/2 = 700 MIPS
– Workloads, jobs, transactions that are CPU bound and very response time or
elapsed time sensitive many not want to exploit the throughput benefits of SMT
• Capacity planning tools for SMT
– For IFLs: Assume 20% increase in throughput due to SMT
– For zIIPs: Assume 25% increase in throughput due to SMT
– Capacity planning tools will initially default to 20% throughput improvement for IFLs and 25% throughput
improvement for zIIPs
– Tools will be updated with the latest information
– Remember that SMT throughput is in addition to benefits seen from the faster uni processor
• Overall guidance: Processor speed + SMT benefit
– IFL: 32% = 10%+ + 20%+
– zIIP: 40% = 10%+ + 25%+
Note: SMT is designed to deliver better overall throughput for many workloads.
Performance in some cases may be superior using single threading
A3 B3 C3
A2 B2 C2
Scalar
SINGLE INSTRUCTION, SINGLE DATA
SIMD
SINGLE INSTRUCTION, MULTIPLE DATA
Instruction is performed for
every data element
Perform instructions on
every element at once
Sum and Store
C1
C2
C3
A1 B1
A2 B2
A3 B3
INSTRUCTION
A1 B1 C1
Sum and Store
SIMD (Single Instruction Multiple Data)
Increased parallelism to enable analytics processing
• Smaller amount of code helps improve execution efficiency
• Process elements in parallel enabling more iterations
• Supports analytics, compression, cryptography, video/imaging processing
• Exploitation by (partial list):
– Java8 and C/C++ for z/OS and Linux on System z; GCC for Linux on System z
– Enterprise COBOL for z/OS, PL/1
– MASS and ATLAS math libraries from Rational for z/OS and Linux on System z
– ILOG-CPLEX, z/OS XML System Services
50
26. 12/10/2015
26
Single Instruction Multiple Data (SIMD)
Introduction and background
51
Motivation / Background
– The amount of data is increasing exponentially - IT shops need to respond to the diversity of data
– Enterprises use traditional integer, floating point data; also string, XML character-based data
– It’s becoming more important for customers to do computations, analytics closer to the data
Customer Perception of Analytics and System z
– System z handles OLTP and Batch; lots of math and changing data is too compute intensive for z
Reality of Analytics and System z
– For last 2-3 generations, z has changed its capabilities in compute processing space (analytics)
• Superscalar, Out of Order (OoO), compiler improvements, floating point
• Legacy Capabilities: quad precision floating point, fuse/multiply/add
– SIMD provides next phase of enhancements for analytics competitiveness on z
SIMD Objective
– Leverage data intensity and be competitive with large data volumes; compete by doing more operations on a given
byte of data, extract more interesting insight, and turn that insight into customer inspiration
Workloads that may benefit from Data Parallelism (SIMD)
– High Data Intensity (i.e. data volume)
– High Compute Intensity (i.e. operations on a given byte of data)
– Predictive IT analytics, Advanced Security/Crypto, BI reporting, Prescriptive Analytics, Next-Gen Data Warehousing
Use Cases
– Reporting functions: Querying and populating reports, often in batch fashion to process lots of data quickly
– Numerically intensive processing
• i.e. time forecasting, simulation; ex: Tivoli based analysis with capacity management
– Modelers, matrix intensive computations
– ILOG CPLEX: optimizations; i.e. delta cruise scheduling, linear programming
Instruction pool Data pool
Results
Instruction pool Data pool
Results
Single Instruction Multiple Data (SIMD)
Vector Processing
• When used with provided libraries and compilers:
– creates a platform for numeric and data intensive computing,
– minimizing effort on the part of middleware/application developers for exploitation.
This capability enhances analytics and big data capability of z13
Key enablers:
Libraries: MASS, ATLAS
Compliers: XLC, Java.Next
String processing:
Java / Cobol / PL1 string, XMLSS, Cognos
Binary Floating Point
ILOG, SPSS, analytics, mobile codes
OS/Hypervisor Support:
z/OS: 2.1 SPE
Linux: SLES12 SP1 and RHEL7.2
Workloads
Java.Next C/C++Compiler built-ins
for SIMD operations
(zOS and zLinux)
MASS & ATLAS
Math Libraries
(zOS and zLinux)
SIMD Registers and Instruction Set
52
27. 12/10/2015
27
SIMD - Exploitation
New z13 assembler instructions which directly use the vector facility (this is not the full list):
• VL Vector Load
• VLL Vector Load with Length
• VSTL Vector Store with Length
• VCEQ Vector Compare
• VFAE Vector Find Any Element Equal
• VFEE Vector Find Element Equal
• Using these instructions promises maximum improvements from SIMD
• ILOG CPLEX, COBOL Inspect .. Tallying, JAVA8, self written ASM programs
All floating point operations will benefit – without change – from the new VFU design (all units
are doubled now compared to zEC12).
tor Element Rotate and Insert under Mask
53
Java 8
SIMD Vector Engine Exploitation
54
28. 12/10/2015
28
Java Performance on z13
By Java version, HW generation, and w/w/o SMT
30 % increase
in Throughput
55
1999
2009
1998
2001
2003
2005
200
7
SDK1.4
1. 31-bit z/OS and 31-bit and 64-bit Linux on z
2. GA 4Q2002
3. z/OS End of Marketing September, 2008
4. z/OS End of Service September, 2011
31-bit and 64-bit SDK 5
1. IBM J9 2.3 VM and JIT Technology
2. z9 Exploitation
3. GA 4Q2005
4. z/OS and Linux on z
31-bit SDK1.1.8
1. OS/390 GA 1999
2. Out of service
31-bit and 64-bit SDK 6 , V6.0.0
1. Supplies Java SE 6 APIs
2. z10 Exploitation
3. IBM J9 2.4 VM and JIT Technology
4. GA 4Q2007
5. z/OS and Linux on z
31-bit SDK1.3.1
1. z/OS and Linux on z
2. GA 3Q2000
3. End of Service: September, 2007
z/OS 64-bit SDK 1.4.2
1. IBM J9 2.2 VM and JIT Technology (1st product use)
2. GA 4Q2004
3. End of Service September, 2008
31-bit SDK1.1.1, then 1.1.4 and 1.1.6
1. First OS/390 Java product – GA 1997
2. Out of service
2014
31-bit and 64-bit z/OS Java SDK 6
V6.0.1
1. Supplies Java SE 6 APIs
2. z196 Exploitation
3. New IBM J9 2.6 VM and JIT Technology
4. Enhanced JZOS and z/OS Security
5. z/OS Java products, GA March 2011:
System z Java Product Timeline
31-bit and 64-bit Java SDK 7.x
1. z/OS and Linux on z
2. Supplies Java SE 7 APIs
3. OpenJDK
4. z196/zEC12 Exploitation
5. New IBM J9 2.6/2.7 VM and JIT
Technology
6. GA Oct 2011
Testimonials: http://www.ibm.com/software/os/systemz/testimonials/
IBM continues to invest aggressively in
Java for System z, demonstrating a rich
history of innovation and performance
improvements.
29. 12/10/2015
29
STI
z990/z890
STI
z9
InfiniBand
z10/z196/z114/
zEC12/zBC12
STI: Self-Timed Interconnect
6 GBps
2 GBps
PCIe Gen2
zEC12/zBC12/
z196/z114
16 GBps
System z I/O Subsystem
Internal Bus Interconnect Speeds
PCIe Gen3
z13
8 GBps
2.7 GBps
57
New FICON Function on z13
• 16 Gbps Link Speeds
– Designed to reduce I/O latency to improve response time for performance-critical middleware and to
shrink the batch window required to accommodate I/O bound batch work
• FICON Dynamic Routing – September 2015
– Designed to allow ISL sharing by FC and FCP traffic to optimize use of ISL bandwidth in the SAN
fabric for both types of traffic
• SAN Fabric Priority – September 2015
– Extends z/OS WLM policy into the SAN fabric
– Gives important work priority to get through SAN traffic congestion (e.g. after SAN hardware
failures)
• zHPF Extended Distance II – June 2015
– Up to 50% I/O service time improvement for remote write
– Designed to help GDPS HyperSwap configurations with secondary DASD in remote site
• 32K devices per FICON channel
– Up to 85 Logical Partitions: More flexibility for server consolidation
• Preserve Virtual WWPNs for NPIV configured FCP channels
– Designed to simplify migration to a new-build z13
• Forward Error Correction Codes – September 2015
– Designed to addresses high bit-error rate on high frequency (>= 8Gb/s) links
– Estimated equivalence to doubling optical signal power
• 6th Logical Channel Subsystem
– Up to 85 Logical Partitions: More flexibility for server consolidation
• 4th Subchannel Set
– Simplifies I/O configurations for a 2nd synchronous copy of data
– With multi-target PPRC, can do HyperSwap and still maintain synchronous copy for 2nd HyperSwap58
30. 12/10/2015
30
DS8870 and z13
Improve performance and resiliency for mainframe environments
59
• January 2015 Preview announcement DS8870
• FICON Dynamic Routing
– Reduce costs with improved and persistent performance for
supporting I/O devices
• 16Gb host adapters
– Improve network performance with 2x faster FC and FICON
adapters; minimize latency for DB2 log writes with zHyperWrite
• Forward Error Correction
– Preserve data integrity with more redundancy on the information
transmitted via 16Gb adapters
• zHPF Extended Distance II
– Increase remote data speed with 50% better IO performance for
remote mirror
• Fabric Priority
– Improved resiliency capabilities while enhancing the value of FICON
Dynamic Routing
1200
14000
31000
20000
52000
23000 23000
92000
110000
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
I/O driver benchmark
I/Os per second
4k block size
Channel 100% utilized
FICON
Express4
and
FICON
Express2
z
H
P
F
FICON
Express8
z
H
P
F
FICON
Express8
FICON
Express4
and
FICON
Express2
ESCON
z
H
P
F
FICON
Express8S
FICON
Express8S
z10 z10
z196
z10
z196
z10
zEC12
zBC12
z196,z114
zEC12
zBC12
z196,z114
350
520
620
770
620 620
1600
2600
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
3000
FICON
Express4
4 Gbps
I/O driver benchmark
MB/second
Full-duplex
Large sequential
read/write mix
FICON
Express4
4 Gbps
FICON
Express8
8 Gbps
FICON
Express8
8 Gbps
FICON
Express8S
8 Gbps
FICON
Express8S
8 Gbps
z10 z10
z196
z10
z196
z10
zEC12
zBC12
z196,z114
z
H
P
F
z
H
P
F
z
H
P
F
zEC12
zBC12
z196,z114
z13
GA1
z
H
P
F
FICON
Express
16S
z13
GA1
FICON
Express
16S
z13
GA1z13
GA1
FICON
Express
16S
16 Gbps
FICON
Express
16S
16 Gbps
z
H
P
F
zHPF and FICON Performance z13 GA1
DS8000 enhanced to support FICON16S
63% increase
60
20% increase
(Controlled measurement environment, results may vary)
31. 12/10/2015
31
60000
84000
92000
110000
0
20000
40000
60000
80000
100000
120000 I/Os per second
Read/writes/mix
4k block size, channel 100% utilized
z10 z196, z10
z13 GA1zEC12
zBC12
z196, z114
520
770
1600
2500
0
500
1000
1500
2000
2500
MegaBytes per second (full-duplex)
Large sequential
Read/write mix
z10 z196, z10
56% increase
20% increase
FE4
4 Gbps
FE8S
8 Gbps
zEC12
zBC12
z196, z114
FE8
8 Gbps
FE8S
8 Gbps
FE16S
16 Gbps
FE16S
16 Gbps
z13 GA1
FE8
8 GbpsFE4
4 Gbps
FCP Performance for z13 GA1
61
FICON Express 16S
• 16 GB/sec link speed
–QLogic HBA supports 4G / 8G / 16G
–New standard 64b / 66b encoding
• Improves link data efficiency
–HBA code in IBM memory
• Simplifies Qlogic code updates
• Reduced latency for large block transfers
–z/OS DB2 Log writes
12-14% latency reduction in 128K log writes
Up to 5-6% improvement in Transaction Latency
–z/OS Managed file transfer
DS8000 zDDB feature - exchange data through the SAN
–Reduced batch window
• Forward Error Correction
–Addresses higher error rates at 16GB
• Sensitivity to damaged / faulty optical cables
–Improved fault isolation for optical links
–IBM leading standards to enable FEC
• Fabric Priority
–Cooperative design across adapter / switch / control unit
–Each operation given priority by z/OS / WLM
• Priority propagated across SAN fabric
–Ensures consistent performance for most important work
Z13 GA1 (1Q2015)
FL
SFP+
SFP+
DRAM
PG2
PG2
CNA
CNA
FL
CFAM-S
SW
Storage Card (LX, SX)
PG2 ASIC
4/8/16GFC CNA ASIC
Universal FC/FICON spare
FICON Express16S Card
62
32. 12/10/2015
32
More Memory Makes a Huge Difference
Adding more memory can enable you to…
– Cut response time – up to 70% reduction for SAP transactions
– Achieve faster decision making by leveraging in memory data
– Shrink batch windows with no change to applications
– Deploy and support more Linux workloads in the same system footprint
– Improve system performance (reduced MLC), minimize constraints, and simplify management of
applications with database exploitation of additional memory
– Get more work done
z13 offers up to a Maximum of 10TB (10,000 GB)
Up to 2.5 TB per model increment
Leadership reliability and availability with RAIM
LPAR support of the full memory configured
Elimination of the 1TB limit
z/OS 2.1 (with SPEs) to support up to 4TB per image
z/VM 6.4 to support up to 1TB per VM guest
63
64
Memory Delivers Additional Benefits
Go for mega memory…
Enable totally new types of applications
− Perform faster table scans with in memory data for faster
response time; reduce CPU by avoiding IO
Simplify Memory capacity planning
− Reduce need to fine tune memory
Accommodate growing batch workloads
− Run sorts using large memory, improving CPU
consumption and elapsed times
Now easily support new modern computing
languages and architectures
− Java and other memory intensive languages
Customers can see CPU savings
− See up to 5% CPU savings with DB2 tuning enabled by
Large Memory
− See 5% CPU savings for typical workloads, in some
cases up to 20% in certain environments, e.g.; when
using SAP with DB2
− Your mileage may vary, and is highly depending on
buffer pool hit ratios
70GB buffer pools
1MB frames with Page Fixed is the best
performer
Candidates benefiting from large
memory include:
Analytics
Java
DB2
Cognos
Indexing
Batch
LE
CF
PR/SM
1.5
1.55
1.6
1.65
4K Pagable 4K Fixed 1M Pagable 1M Fixed
milli-seconds
Total DB2 CPU time per Transaction
33. 12/10/2015
33
z13 PR/SM Role and Capabilities expanded
Dynamic Memory Management!
• zEC12 and earlier PR/SM controlled
relationship between logical and physical
CPs. it pseudo-dedicate physical CPs to
vertical high logical CPs, and try to keep
logical CPs for an LPAR together in 1 book
and one chip, to maximize the value of the
L3 and L4 cache. PR/SM can change
relationship and transparently move logical
CPs between physical CPs.
• z13 capabilities of PR/SM expands. Not only
assigning logical CPs to physical CPs, now
also responsible for controlling where the
memory for each LPAR will be allocated.
And it is now able to not only dynamically
move logical CPs between drawers, it can
also (transparently) move memory between
drawers.
• The complex algorithms PR/SM uses to
determine best location for each LPAR
include considerations, including:
– The number and type of logical CPs
in LPAR.
– The LPAR's weight as % of total
weight of all active LPARs (fair
share' of overall capacity).
– The number of vertical high, vertical
medium, and vertical low (“parked”
and “un-parked”) CPs in the LPAR.
– The amount of LPAR memory and
drawer memory.
– The number of assigned and
unassigned PUs in each chip, node,
and drawer.
– Many others.
65
IBM z13 Large Memory
Client Value from Large Memory
• Response Time: Consistent, fast transactional response time drives a series of
bottom line benefits for Clients including higher productivity and sales.
• Availability Gains: The culture to hoard memory and tune tightly exists. 3X and
mega memory help break this. Availability gains show up in (1) how well the system
survives I/O disruptions (2) workload tuning to handle spikes and (3) all of the Flash
use cases get better (workload startup and failure scenarios).
• Application Productivity and Memory Hungry Workloads: Analytics and other
large memory users can be placed closer to data. Changes to charge back
allocation reflecting the lower cost of memory enable Application Designers' use of
caching software layers and other industry performance and productivity technology.
• Long Term Architectural Vision: The 3X memory purchase acts as a shock to the
system to move the balance point of workloads towards gaining the advantages of
large memory usage. This change in vision in an organization is a very significant
Client advantage of the 3X and Mega memory pricing approaches.
(Controlled measurement environment, results may vary)
http://www.redbooks.ibm.com/abstracts/redp5146.html?Open
34. 12/10/2015
34
IBM z13 Large Memory
Enabling optimal middleware efficiency
• DB2 Buffer pools
– By configuring more memory for DB2 buffer pools
• up to 3% CPU time savings
• up to 10% transaction response time savings in the DB2 portion of their workload
– DB2 synchronous I/O simulation tool now available on DB2 11
• Predict expected reduction in sync I/O resulting from a specific increase in the size of a
DB2 buffer pool
• Estimate CPU savings based on reduction in sync IOs /sec
• IMS
– Page fix IMS Full Function database buffers
• Improve response time for IMS DB accesses from application
• Save up to 3% CPU time of the IMS portion of your workload
• MQ
– Large memory for IBM MQ V8 can help to cost effectively manage the
increasing message volumes generated from today's mobile and cloud
applications
(Controlled measurement environment, results may vary)http://www.redbooks.ibm.com/abstracts/redp5146.html?Open
IBM z13 Large Memory
Unleashing the Next Wave of Innovation
• MDM (InfoSphere)
– Right-sizing your group buffer pools
• Configure large enough group buffer pools to cache all MDM Tables and indexes
– Internal measurements
• Increasing group buffer pool size from 120 GB to 565 GB and retaining all MDM
tables and indexes in the group buffer pool
• Reduced CPU time by 6.8%
• Reduced application elapsed time by 18%
• Java / WAS
– Allow larger Java heaps and more Java instances
• Reduce latency and CPU cost
• Cognos Dynamic Cubes on z/OS
– Data in memory for faster report generation
– Scale up from 1 to 3 Cognos instances with
– Increasing one COGNOS Instance to 3 instances using 480GB, we
measured 95% scale efficiency
(Controlled measurement environment, results may vary)http://www.redbooks.ibm.com/abstracts/redp5146.html?Open
35. 12/10/2015
35
IBM z13 Large Memory
Demonstrating Customer Value
• Ensure that middleware is optimally tuned
– Are DB2 buffer pools optimized?
• Are the DB2 buffer pools sized correctly?
– Customers with DB2 V11 to use DB2 synchronous I/O simulation tool.
• Is the customer using fixed large pages for buffer pools?
– Is other middleware taking full use of memory (IMS, MQ, MDM)?
• There are other opportunities to exploit mega-memory
– Benefits from more JVMs, or larger JVM heaps?
• WAS applications, Java applications, Java transactions
– Other opportunities?
http://www.redbooks.ibm.com/abstracts/redp5146.html?Open
Large Memory remarks
Rethink memory usage on z Systems
• With z13 option to include up to 10 TB in the machine, 2.5 TB per drawer.
– “Mega memory“ affordable option 1H 2015
• Large memory expolitation?
– Assiging the memory to Linux images, as Linux applications often like that.
– Assigning memory to DB2 Bufferpools. Good idea, also consider:
• Adding more memory to local buffer pools means you need to add memory in the
global bufferpools (CF)
• If your customer already has high local bufferpol hit ratios, larger bufferpools offer
less value. Consider revamping BP layout.
– Other middleware smaller benefits from large memory than DB2 (like IMS, CICS, ...)
– Java may use large memory
– For MQ Series, Version 8 provides 64 bit support
• Application programmers should be involved in planning for large memory
exploitaion.
36. 12/10/2015
36
Large Memory Measurement Results
DB2 and SAP
http://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP102461
z13 Extends Scale, Relieves Constraints
Advanced system design optimized for digital business
* Servers exploit a subset of its
designed I/O capability
PCI – Processor Capacity Index
Customer
Processors
PCI for
1-way
1695
Memory
80-way
64-way
54-way
1.5
TB
512
GB
1202920600
288 GB/sec*
172.8 GB/sec*
15033
TB
384 GB/Sec*
101-way
z10 EC
z9 EC
z196
zEC12
z13
10
TB
141-way
System I/O Bandwidth
832 GB/Sec*
72
37. 12/10/2015
37
Acceleration & Optimization
Integration versus time-to-market,
• Embedded accelerators
• Computation engines for analytics
• Assists for dynamic software optimization
• Enablement for integrated function
• Close collaboration with compiler and other software teams
• PCIe-attached accelerators
• FPGAs, Field Programmable Gate Arrays
• Leverage flexibility for special functions
• In-line processing of data entering or leaving system
• Off-load specialized data processing
• Crypto, FICON, Compression, ROCE, Flash,...
• More may come...
• Heterogeneous system optimization
• Integrate special-purpose appliances into System z workloads
• Enhanced system scale and price/performance
73
IBM z13 Designed for Analytics - Summary
Accelerate insight and simplify implementation
• IBM DB2 Analytics Accelerator accelerates
queries for faster insight
– New innovative use cases, such as in-
database transformation and advanced
predictive analytics
• Large memory allows new opportunities for
in-memory computing
– per system & per LPAR
• SMT2 for increased zIIP & IFL cores
capacity
• SIMD delivers accelerated analytics
processing for complex queries
– Enable vector processing capabilities to z
Systems
• More
– Optimized math libraries and compilers that
will speed up and simplify application
development
– Faster thread speeds
– z Enterprise Data Compression (zEDC) to
improve the economics of keeping data on
z Systems
74
Which approach is designed
for the highest volume of
traffic? Which road is faster?
Illustrative numbers only
A3 B3
C3
A2 B2
C2
Scalar
SINGLE INSTRUCTION, SINGLE DATA
SIMD
SINGLE INSTRUCTION, MULTIPLE DATA
Instruction is performed for
every data element
Perform instructions on
every element at once
Sum and Store
C1
C2
C3
A1 B1
A2 B2
A3 B3
INSTRUCTION
A1 B1
C1
Sum and Store
38. 12/10/2015
38
Analytics as part of the flow of business; Insights on every transaction
• Purchase made
• Resources
consumed
• Bill paid
• Claim submitted
• Information
updated
• Call center
contacted
• What happened?
• How many, how
often, where?
• What actions are
needed?
• What will happen if?
• What will produce
the best outcome?
Transactions & Analytics processed together
Made possible by using LARGE memory - in addition to SIMD and SMT
75
What’s changing?
Data is the new basis of competitive advantage
Data is the
world’s
newest
resource
Decision-making
extends from
few to many
As data value
grows, current
systems won’t
keep pace
76
39. 12/10/2015
39
77
What is Big data?
Google can give you nearly 2 Billion options
Vendors have even more definitions
Here is how Gartner defines Big Data
Big data is high-volume, high-velocity and high-variety information
assets that demand cost-effective, innovative information
processing for enhanced insight and decision making.
Big Data Growth is Astounding…
The four V’s of Big Data
1 in 2
business leaders don’’’’t
have access to data they
need
83%
of CIO’’’’s cited BI and
analytics as part of their
visionary plan
54%
of companies use analytics
for competitive advantage
80%
of the world’s data
today is unstructured
90%
of the world’s data was
created in the last two
years
20%
is the amount of
available data traditional
systems leverages
Source: GigaOM, Software Group, IBM Institute for Business Value“, “Volume, Velocity, Variety, Veracity”
78
40. 12/10/2015
40
Extracting insight from an immense volume, variety and velocity of data, in context, beyond what
was previously possible
Big Data
includes any of the following Characteristics
Manage the complexity of data in
many different structures, ranging
from relational, to logs, to raw text
Streaming data and large volume
data movement
Scale from Terabytes to
Petabytes (1K TBs) to Zettabytes
(1B TBs)
Establish trust as the number of
data sources grows
Variety:
Velocity:
Volume:
Veracity:
Create IT AgilityManage RiskOutperform
Why Act Now?
Only 1 in 5 organizations
allocate more than 50% of
IT budget to new projects
Of leaders cite growth as
the key source of value
from analytics
Source:
1 - IBM IBV Study: Analytics: A blueprint for value, October 2013
2 - IBM Global Study on the Economic Impact of IT Risk, 2013
3 - IBM Global Data Center Study, 2012
Of respondents were impacted
by a cyber security breach
over the past 24 months
46%75% 1in5
41. 12/10/2015
41
Analytics is becoming the Keystone of every
organization …
• Analytics derive insight from data
– To help optimize business performance
– To build new innovative services
– To fight against fraud
– To make all customer interaction personal! …
• Analytics become Business Critical!
– Analytics services are tightly integrated with business critical applications and data
• Often hosted in z/OS transaction and batch systems
• Often relying on copies or aggregation of transaction and application data
– Analytics is part of the flow of the business.
– Decision processes have to be improved with new business insight derived from real
time or near real time data.
– Failure of these applications for any length of time can result in lost business or
reputation.
– Analytics solutions need to support a large concurrent user population with high volumes
of requests.
• Analytics are only as good as the underlying data foundation
– Data governance & Security & Performance
81
… but IT remains aligned to the old way
of doing business analytics.
• Some reluctances from the past
– Core business is primary, analytics is secondary!
• On core business side: High volume transactions and batch processing running concurrently,
shared everything DB
• On analytics side: Low volume complex queries and batch reporting, shared nothing DB
– Cost of running analytics on z … without looking at all hidden costs concerning data
movement – latency, data governance, IT complexity
– Impact on operational performance & security
• Key drivers to change IT perception
– Awareness of z position as primary Systems of Records
– Technology availability to build a fully-integrated, end-to-end system that executes
intelligent business processes
– Recognized business value of advanced real time analytics
– Business leaders brainstorm to identify to rethink business process with HTAP influence
• Gartner Research Note G00259033 28 January 2014: Hybrid Transaction/Analytical Processing
Will Foster Opportunities for Dramatic Business Innovation
Insights on every
transaction
Analytics as part of the flow
of business
Analytics close to the data
z Systems
Transactions
Analytics
8282
42. 12/10/2015
42
Data
Moved
Analytics Approaches for Mainframe Data
DATA
Transaction Data
Customer Data
Account Data
Payment Data
Claims Data
System z Host
System z/OS
Systems of Record
Orchestrate
Processing Predictive
Scoring
Business
Rules
Move the data to
the analytics
Performance of critical transactions may not meet
SLAs due to data movement
Customer needs to create security
infrastructures across multiple servers
Customer needs to create audit infrastructure
across multiple servers to ensure governance
Customer needs to create availability and DR
function for multi-server transaction flows and in-
transit data
Move the analytics to
the data, and within
the transaction
Unparalleled, proven performance execution for
models
and rules, with NO or seamless data movement
Leverage existing best of class security with
System z infrastructure
Leverage existing transaction level auditing
and logging for governance
Leverage existing, tested HA / DR capabilities
already configured with System z
Network
Network
Data
Moved
Data
Moved
Network
DATA
Transaction
Data
Customer
Data
Account Data
Payment Data
Claims Data
Orchestrate
Processing
Predictive
Scoring
Business
Rules
System z/OS
Systems of Record
System z Host
83
System z Mainframe
CP(s)
z/OS
IFL
…
z/VM
S
MF
Linux for System z
IBM InfoSphere BigInsights
HDFS
MapReduce, Hbase, Hive
IFL IFL
DB2
VSAM
IMS
Logs
System z
Connector
For Hadoop
System z
Connector
For Hadoop
Simplified data transfer from z/OS to Hadoop
IBM InfoSphere System z Connector for Hadoop
Easily extract data from
mainframe sources
Purpose built for Hadoop
Complementary to other IBM
tools: IBM MQ FTE, DataStage
Fast, secure data transfer
Interactive or batch
Supports popular Hadoop
distributions
IBM BigInsights
Cloudera CDH
Hortonworks HDP
43. 12/10/2015
43
Operational
Data Store
Enterprise
Data
Warehouse
Analytics
Accelerator
What is
happening?
What
happened?
What is likely
to happen and
what do I do
about it?
OLTP Transactions
Operational analytics
Real time data ingestion
High concurrency
Advanced analytics
Standard reports
Complex queries
Historical queries
OLAP
Integrated Transformation/Warehousing
Single DB2 z/OS Data Sharing Group
Big Data
Accelerator
Accelerated Reporting
Real-Time Predictive and Prescriptive Analytics
Customer
Interaction
Data In
Business
Insight Out
z Systems Hybrid Transaction
and Analytic Processing
Everything is online – Analytics in the Right Place
85
Why did it
happen?
Hybrid transaction/Analytical processing
The hybrid computing
platform on z Systems
Supports transaction processing
and analytics workloads
concurrently, efficiently and
cost-effectively
Delivers industry leading
performance for mixed workloads
The unique heterogeneous scale-
out platform in the industry
Superior availability, reliability
and security
Transaction
Processing
Analytics
Workload
44. 12/10/2015
44
• Business agility through simplified architecture with in-database
transformation and multi-step processing
• Real-time, actionable business processes through in-database analytics
• Insight into now to maximize business opportunities through enterprise
Incremental Update enhancements
Announcing DB2 Analytics Accelerator Version 5.1
Enabling real-time analytic solutions on a single, integrated system combining
transactional data, historical data and predictive analytics
• Extended security through encryption of data at rest
and in motion while taking advantage of the renowned
built-in security of z Systems
• Enriched systems management capabilities and
improved serviceability through IBM Call Home
Strategy
Enable DB2 transition into a truly universal DBMS that provides best
characteristics for both OLTP and analytical workloads.
Complement DB2's industry leading transactional
processing capabilities
Provide specialized access path for data
intensive queries
Enable real and near-real time analytics
processing
Execute transparency to the applications
Operate as an integral part of DB2 and z
Systems
Reusing industry leading PDA's query and
analytics capabilities and take advantage of
future enhancements
Extend query acceleration to new, innovative
usage cases, such as:
– in-database transformations
– advanced analytical capabilities
– multi-temperature and storage saving
solutions
Ultimately allow consolidation and unification of transactional and
analytical data stores
DB2 for
z/OS
In-database
Transformation
Query
Accelerator
Storage
Saver
OLTP
Advanced
Analytics
45. 12/10/2015
45
IBM DB2 Analytics Accelerator
Do things you could never do before!
• What is it?
– The IBM DB2 Analytics Accelerator is a workload optimized, appliance add-
on to DB2 for z/OS, that enables the integration of analytic insights into
operational processes to drive business critical analytics and exceptional
business value
• What does it do?
– Accelerates complex queries, up to 2000x faster
– Lowers the cost of storing, managing and
processing historical data
– Minimizes latency
– Reduces z Systems capacity requirements
– Improves security and governance
– Reduces operational costs and risk
– Complements existing investments
89
IBM DB2 Analytics Accelerator
Product Components
OSA-
Express4
10 GbE
CLIENT
Data Studio
Foundation
DB2 Analytics
Accelerator Admin
Plug-in
z Systems
Data Warehouse application
DB2 for z/OS enabled for IBM
DB2 Analytics Accelerator
IBM DB2
Analytics
Accelerator
Netezza
Technology (PDA)
Users/
Applications
Note: There are several connection options using switches to increase redundancy
Dedicated highly available
network connection
90
46. 12/10/2015
46
Customer Table
~ 5 Billion Rows
300 Mixed Workload Queries
Times
Faster
Query
Total
Rows
Reviewed
Total
Rows
Returned Hours Sec(s) Hours Sec(s)
Query 1 2,813,571 853,320 2:39 9,540 0.0 5 1,908
Query 2 2,813,571 585,780 2:16 8,220 0.0 5 1,644
Query 3 8,260,214 274 1:16 4,560 0.0 6 760
Query 4 2,813,571 601,197 1:08 4,080 0.0 5 816
Query 5 3,422,765 508 0:57 4,080 0.0 70 58
Query 6 4,290,648 165 0:53 3,180 0.0 6 530
Query 7 361,521 58,236 0:51 3,120 0.0 4 780
Query 8 3,425.29 724 0:44 2,640 0.0 2 1,320
Query 9 4,130,107 137 0:42 2,520 0.1 193 13
DB2 Only
DB2 with
IDAA
270 of the Mixed
Workload Queries
Executes in DB2 returning
results in seconds or sub-
seconds
30 of the Mixed Workload Queries took minutes to hours
Successfully accelerated the problem queries without affecting the rest
Customer “A” Example
91
Without
accelerator
With
accelerator
Customer Example –
LPAR CPU utilization comparison with and without IDAA
92
47. 12/10/2015
47
Analytic needs are expanding from
enterprise data to big data
5 Facts to consider to ensure success
1 Organizations are using analytics to outperform their competition
2 More users across the organization want access to analytics,
woven into the fabric of the business
3 Analytics is business critical and demands low latency, high
qualities of service and performance
4 Spreading analytic components across multiple departments can
increase data latency, cost, complexity and governance risk
5
Bringing analytic components to where data originates improves
data governance while minimizing data latency, cost and
complexity
93
Five Business Critical Analytics Use Cases
Big Data Exploration
Find, visualize, understand
all big data to improve
business knowledge
Enhanced 360o View
of the Customer
Achieve a true unified view,
incorporating internal and
external sources
Operations Insight
Analyze a variety of machine
data for improved business results.
Data Warehouse Augmentation
Integrate big data and data warehouse
capabilities to increase operational efficiency
Security/Intelligence Extension
Risk, compliance and counter fraud
detection. z13 can deliver real time
analytics and scale up to meet future
demand
94
48. 12/10/2015
48
More Users across the Organization
want access to Business Critical Analytics
95
The “Big Data” Starting Point
Where are organizations getting the most return on Big Data projects?
0 10 20 30 40 50 60 70 80
Audio
Video
Others
Images
Geospatial data
Free-form text
Social media data
E-mails or documents
Machine or sensor data
Log data
Transactions
Percentage of
respondents
Gartner research note “Survey Analysis - Big Data Adoption in 2013 Shows Substance Behind the Hype“ Sept 12 2013
Analyst(s): Lisa Kart, Nick Heudecker, Frank Buytendijk
N=465, multiple responses allowed
Most big data initiatives
involve transactional or log-file
data
96
49. 12/10/2015
49
Unfortunately for most of our clients, their data lifecycle is too
fragmented to gain advantage from that data
• Significant complexity
– Data is move from operational databases
to separated data warehouses/data marts
to support analytics
• Analytics latency
– Transactional data is not readily or easily
available for analytics when created
• Lack of synchronization
– Data is not easily aggregated and users
are not assured they have access to
“fresh” data
• Data duplication
– Multiple copies of the same data is
proliferated throughout the organization
• Excessive costs
– An IT infrastructure that was not designed
nor can support real-time analytics
Historical
Data
Predictive
Data
Transaction
Data
97
Business Critical Analytics Systems with IBM z Systems
An Hybrid Vision
Bring analytics to the data
Reduced latency
Reduced complexity
Reduced cost
Deliver business critical analytics
Timely, accurate, secure data
Availability, scalability,
performance
Rapid deployment & expansion
Evolve with the business
Start with your top analytic
requirement(s)
Grow without changing customer
existing IT environment
Business
System
OLTP & Batch
Business
Critical
Analytics
Improved business
performance out
Transactions in
Data
Transformati
on
Data
Warehousing
Minimize latency. Improve performance. Drive innovation.
• Purchase made
• Resources consumed
• Bill paid
• Claim submitted
• Information updated
• Call center contacted
• What happened?
• How many, how often, where?
• What actions are needed?
• What will happen if?
• What will produce the best
outcome?
98
50. 12/10/2015
50
On average, 70% of the data that feeds data warehousing and
business analytics solutions originates on the System z platform
(financial information, customer lists, personal records,
manufacturing…)
DB2 Analytics Accelerator – Four Usage
Scenarios
Understand your workload and data:
Where transaction source data
is being analyzed today
Use Case Benefits
If the data is analyzed on the
mainframe
Rapid Acceleration of
Business Critical Queries
Performance improvements and cost
reduction while retaining System z
security and reliability
If the data is offloaded to a distributed
data warehouse or data mart
Reduce IT Sprawl for
analytics
Simplify and consolidate complex
infrastructures, low latency, reliability,
security and TCO
If the data is not being analyzed yet
Derive business insight from
z/OS transaction systems
One integrated, hybrid platform,
optimized to run mixed workload.
Simplicity and time to value
If the analysis is based on a lot of
historical data
Improve access to historical
data and lower storage costs
Performance improvements and cost
reduction
1
2
3
4
99
System z Point of View
Building a foundation to grow with business needs
10
0
Why z13?
3X larger memory enables in-memory analytics for faster insight
MASS libraries can see a 2X to 10X improvement making it
advantageous to port x86 analytic workloads
CPLEX on z/OS exploitation of SIMD instructions provides up to 80%
improvement complex modeling
Add real-time scoring to your OLTP workload with minimal impact on
CPU consumption
zIIP exploitation of SMT2
Linux exploitation of SMT2
Why System z
Currency of data
Reduce complexity
Bring analytic function to the data
Improve synchronization
Eliminate data duplication
51. 12/10/2015
51
101
Systems of Record
Systems of Record are well integrated
and mostly complete
Order
Fulfillment
Corporate
Data
Ware-
house
Accounting
Finance
z/OS
Systems of Engagement
Systems of Engagement are
disconnected, piece parts
Internet
of Things
(e.g., RFID)
Mobile
Apps
Social
Channels
Siloed
Dept.
Apps
SaaS/
Cloud
Apps
Linux on System z
Campaign Mgmt App
Other Dept.
Apps
System z Bridging
Systems of Record with Systems of Engagement
10
2
z13 Potential Cost Reduction
Use Cases: HW/SW “currency” Summary
Workload Options
Performance
Improvement
(estimation)
Potential Cost
Reduction
(estimation)
SAP with DB2 on z/OS
zHW+zSW currency
+ more memory
10%-18%
+20%
10%
30%
CICS/DB2 banking
zHW+zSW currency
+ more memory
10%-15%
+5%
10%
15%
WAS on z/OS
e.g. internet banking
zHW+zSW currency
+ SMT
+ more memory
10%-12%
+25%
+5%
5%
5%
15%
SPSS
predictive analytics
zHW+zSW currency
+ more memory
+ SIMD
10%-18%
+5%
tbd
10%
15%
tbd
Traditional Batch processing
zHW+zSW currency
+ more memory
+ SIMD
10%
+5%
tbd
10%
15%
tbd
Linux consolidation on IFL
e.g. Private Cloud
zHW+zSW currency
+ SMT
+ more memory
+ GDPS
15%
+20%
tbd
D/R improvement
15%
35%
tbd
tbd
Mobile
zHW+zSW currency
+ more memory
10%
+5%
5%
15%
All performance information was determined in a controlled environment. Results may vary.
Sample output from new version of
z13 Benefits Estimator tool
52. 12/10/2015
52
L1 miss
Instrs
1
2
3
4
5
6
7
Time
In-order core execution
L1 miss
Time
z196 Out-of-order core execution
L1 miss
zEC12 Out-of-order core execution
Time
Improved
overlapping
opportunities
Execution
Storage access
Dependency
Out of Order Execution
z10, z196 vs zEC12/zBC12
12/10/2015
zBC12/zEC12 Cache Topology
L1 64KI + 96KD
8w (D) / 4w (I) Set Associative
L2 Private 1+1MB Inclusive of L1s;
new split 2nd level cache design
L3 Shared 48MB Inclusive of L2s
12w Set Associative
L4 384/192MB Inclusive
24w Set Associative
zBC12/zEC12
4 L4 Caches
384/192MB
Shared eDRAM L4
L1 64KI + 128KD
8w (D) / 4w (I) Set Associative
L2 Private 1.5MB Inclusive of L1s
12w Set Associative
L3 Shared 24MB Inclusive of L2s
12w Set Associative
L4 192MB Inclusive
24w Set Associativez114/z196
4 L4 Caches
192MB
Shared eDRAM L4
6 L3s,
24 L1 / L2s
L2
L1
24MB Shared
eDRAM L3
L2
L1
L2
L1
L2
L1
L2
L1
24MB Shared
eDRAM L3
L2
L1
L2
L1
L2
L1
L2
L1
48MB Shared
eDRAM L3
L2
L1
L2
L1
L2
L1
L2
L1
L2
L1
L2
L1
48MB Shared
eDRAM L3
L2
L1
L2
L1
L2
L1
L2
L1
L2
L1
6 L3s,
36 L1 / L2s
53. 12/10/2015
53
Data
Compression
Acceleration
High Speed
Communication
Fabric
Proactive
Systems Health
Analytics
SSD Flash
Exploitation
Hybrid
Computing
Enhancements
Reduce CP
consumption,
free up storage
& speed cross
platform data
exchange
Optimize server to
server networking
with reduced
latency and lower
CPU overhead
Increase
availability by
detecting unusual
system behavior
for faster problem
determination and
resolution
Improve availability
at critical times like
market open or
during abnormal
situations; New
coupling & Linux
exploitation
x86 blade resource
optimization; New
alert & notification for
blade virtual servers;
Expanding future
roadmap
zEDC
Express
10GbE
RoCE Express
IBM
zAware
IBM
Flash Express
zBX Mod 003;
zManager
Automate; EAM
zEnterprise compilers (COBOL, PL/I, C/C++), Run time monitoring and Transactional Memory provide an
optimized application infrastructure for increased software performance
Innovations available on zBC12 and zEC12
w zEC12/
zBC12
w zEC12
zBC12
w zEC12
zBC12
w zEC12
zBC12
General Performance Disclaimers
MSU
Ratings
MIPS
Tables
IBM System z Capacity Planning in a nutshell
Please do not use “single-number tables” for capacity comparisons
54. 12/10/2015
54
General Performance Disclaimers
IBM System z Capacity Planning in a nutshell
Please do not use “single-number tables” for capacity comparisons
Preparing for z13: Why CPU MF Important?
Collect SMF 113, Calculate RNI, use zPCR
• z13 provides lower single thread improvements than previous processor
changes, e.g. zEC12 versus z196
• z13 provides more variability in capacity improvement
– Capacity projections and expectations should be reasonably accurate
– Relative Nest Intensity (RNI) is a metric describing access to various cache
levels of the processor architecture
55. 12/10/2015
55
Although System z servers are famous for their scalability, CPU consumed/Trans
is a function of total load on system as this real life Nordic client case illustrates
Approx. 3-4% growth in CPU/transaction for each 10% growth in CPU busy
Increasing Data and Compute Requirements
11
0
Mathematical
Optimization
Next Best Action
What will happen
next?
Business
Intelligence-
Reporting
Big Sort
Small
Sort
Small
String
Ops
Large
String
Ops
Big Scan
Traditional IT
Analytics
Predictive IT
Analytics
Data Intensity
Small
Matrix
Math
Large
Matrix
Math
ComputeIntensity
Integer
Ops
Floating-
Point Ops
String
Ops
FP,
String,
Int
String,
Int
Traditional
Algorithms
Competitive
Algorithms
Data
Warehousing
(DW)
Next-
Generation
DW
Small
Compress
Large
Compress
Security/
Crypto
Adv
Security/
Crypto
57. 12/10/2015
57
To MIPS or not to MIPS
that is z question
• You can not/should not just convert MIPS from zEC12 (or older machines) to z13. Hollistic“
view performance needed to make a prediction on what machine you need.
• Collecting data from existing CEC is „strongly recommended“
– HW Instrumentation Services (HIS) is support built into the z/OS operating system to allow
customers to capture CPU MF information, both counter data and sample data.
– Please use the tool from Patrice Megard to analyse the data.
– Use the HIS data also as „feed“ for zPCR.
– This is also important to have a detailed documentation about the status quo before migrating to a
new machine.
• Native Linux does neither support SMT2 nor SIMD until 3Q 2015. Please keep that in mind
when sizing machines for Linux for system z workloads.
• You need to involve Software People in the performance discussions. Their job is to do an
assessment for (list is incomplete):
– Java 8 improvements due to SIMD.
– Cobol 5.2 Compiler improvements and project evaluation with the client.
– Large memory analysis and exploitation activity definition with the client.
– Check that SMT can be exploited as planned.
z13 Sizing Methodology
1. Collect HIS data from the peak hours of the last month (Software AND Hardware)
Please use the IBM supplied tool to analyse the data.
Use the HIS data also as „feed“ for zPCR.
Collect the SMF70 records from the same time intervall
2. Collect CP3000 data from one full week (incl. WE).
It must be the week where the 4HRA is also included (you get this from the SCRT
report) -> Ask your SW rep to provide it to you.
Once you have the data you need, analyse what workload is running in the 4HRA and
what effect a migration to z13 would have to this particular workload.
3. Use zPCR to get a first proposal of the new HW size. This is dependent on the HW
PEAK you have – not the SW peak (the SW peak is a 4hour rolling AVERAGE).
For a first iteration on sizing, leave out SMT benefits.
4. After you analyzed the HIS and zCP3000 data do a workshop with IBM about z13 and its
capabilities and dependencies.
Make sure you involve Software IBMers as they need to explain Java, Cobol, PL1 etc.
Improvements of z13.
58. 12/10/2015
58
2001 20052000 2008 2010 20152003
Capacity comparison based on LSPR & IBM zPCR tool
* Average SMT benefit on z13
Linux Performance Evolution
TCO improvement through server generations, and SODs
+58%
+10-12%
+33%
+61%
+36%
20152012
+20-32%*
+26%
1 IFL on all Systems = 120 PVUs
KVM support for Linux on System z SODGDPS support for Linux on System z SOD (1H 2015)
Source: 2014 IBM Market Intelligence, Percentage of survey respondents
Recommended Linux Workloads
for Linux on z Systems
• Data services: Cognos®, SPSS®, DB2®, InfoSphere™,
Informix®, Oracle Database, IBI WebFOCUS, …
• Business applications: WebSphere Application Server,
WebSphere Process Server, Oracle Application Server, …
• Development and test: WebSphere®/Java applications –
Rational® Asset Manager, Build Forge®, ClearCase®, Quality
Manager
• Email and collaboration: IBM Domino®, IBM Connections, IBM
SameTime, WebSphere Portal, …
• Enterprise Content Management: FileNet® Content Manager,
Content Manager, Content Manager On Demand
• Business Process Management: Business Process Manager,
WebSphere Business Monitor, FileNet Business Process
Manager, WebSphere Operational Decision Management, …
• Infrastructure services: WebSphere MQ, WebSphere Message
Broker, WebSphere Enterprise Service Bus, DB2 Connect™,
FTP, NFS, DNS, Firewall, Proxy, …
• Cloud management: Infrastructure (IaaS), Platform (PaaS),
Software (SaaS), Business Process as a Service – Tivoli®
System Automation Manager, Tivoli Provisioning Manager,
Integrated Service Management for System z, Maximo® Asset
Management, …
• Print (Ubiquitech)
59. 12/10/2015
59
Before
Enterprise
Linux
Servers
2*15 IFL = 30 Oracle RAC EE Licenses
380 cores = 285 Oracle RAC EE Licensesfrom
to
Nordic customer - Oracle consolidation
SOD:
zKVM – An Open Hypervisor for zEnterprise
Open
• Open Source based Virtualization
• Open Source with Enterprise scale capabilities
• Accelerate adoption Linux on System z
Cloud
• Standards-based Cloud enablement
• OpenStack
Efficient
• KVM already used by existing users by FIEs and
MSPs
Use the tools you know and use today
• Puppet, Chef, Heat, Knife, Moab
• Home Grown Scripts in Perl , Ruby , Java….
KVM –
Optimized
for System z
System z Host
z CPU, Memory and IO
Support Element
PR/SM™
. . .
z/TFP
z/OSz/OS
LinuxonSysz
z/OSz/OS
LinuxonSysz
LinuxonSysz
LinuxonSys
z
KVMz/VM
Modern
Open Standard
Simple
LinuxonSysz
A new non
disruptive
hypervisor choice
for the mainframe
z13,
zEC12, zBC12
supported
KVM as an additional choice
to run existing and new Linux
centric workload on
zEnterprise in parallel to your
existing z/VM virtualization
environment and z/OS
11
8
60. 12/10/2015
60
Introducing the z13
Removing constraints on growth through innovation
Greater Workload Performance,
Capacity and Scale in the same
energy footprint
40% more total capacity
40% more configurable cores (up
to 141 vs 101)
New vector facility (SIMD) for
faster mathematical computation
Up to 6 instructions per cycle
(double that of zEC12)
3+ x more memory to reduce
latency (10 TB vs 3 TB)
New multithreading (SMT2) to
expand IFL (Linux) and zIIP
capacity
40% more LPARs to securely host
more cloud tenants (85 vs 60)
z13: An innovative, intelligent and integrated system that provides a trusted
foundation for sustainable growth today and in the future
Better Economics, Flexibility
and Efficiency
Almost 50% increase in
granularity to fine tune system
usage and cost
4x data access with zEDC
Standalone zBX support for
more flexibility
New resilient IO Infrastructure
addresses Skills, Complexity,
Cost and Availability
Price Performance gains for
Linux, zIIP, mobile and new SW
workloads
Investment protection with full
upgradeability from z196 and
zEC12
Resilient and Secure Growth
Highest level of Security (PR/SM EAL5+)
Next Generation Hardware Cryptography
Best System z RAS with integrated sparing
Much enhanced hypervisor
11
9
Focused on Enterprise Linux
Extending Linux to wider audience
with LinuxONE, Linux/KVM (SOD)*
Continuous data availability for z/OS
and Linux guests under z/VM with
new GDPS Appliance (SOD)*
Faster diagnosis with IBM zAware –
now extended to Linux on z Systems
IBM zAware Support for Linux on System z
12
0
• Linux on System z system logs can now be analyzed by IBM zAware
• Upgraded analytics engine for better results on z/OS analysis
• Upgraded internal database for improved RAS
• Completely rewritten UI, including heat map views
HiperSockets ™
OSA (for data from other
servers)
LPAR
z13 Host 1
IBM zAware
Partition
Web Server
Analytics
z/OS
operlog
LOGGER
Data
Transport
Linux on System z
HiperSockets ™
OSA (for data from other
servers)
LPAR
zServer Host 2
z/OS
operlog
LOGGER
Data
Transport
z/OS
operlog
LOGGER
Data
Transport
syslog
Results
Models
Data
Retrieval
Manage zAware
Firmware partition
(similar to CF)
File
System
IBM zAware GUI
Persistent
Storage
Control IBM
zAware-specific
knobs
View
IBM zAware
results
SE
zAware Partition
Shipped as firmware with z13
zVM
61. 12/10/2015
61
z Aware Heat Map – All systems in a group
enhanced zAware interface
z Systems z13 Model Overview
Processors
CP, IFL, zIIP, ICF, optional SAP Specialty engines:
‒ IFL and zIIP exploitation of SMT-2
‒ 2:1 zIIP to CP ratio on a core basis
‒ zAAP eliminated as per SOD
3 Sub capacity levels for up to a 30 way
Full support of Capacity on Demand Features
85 LPARs, increased from 60
Memory
Maximum 10 TB / 2.5 TB per drawer RAIM
LPAR support of the full memory enabled
I/O
Up to 5 PCIe I/O drawers
‒ FICON Exp8S, OSA-Exp4S, OSA Exp5S, 10Gb RoCE,
‒ FICON 16S
Up to 2 Legacy I/O Drawers
‒ Up to16 FICON 8 features
‒ Carry forward only
Crypto Exp4S, Crypto Exp5S,
zEDC,
zFlash Express
Environmental
Enhanced integrated sparing
Directional setting of rear exhaust airflow
1 U rack mountable HMC option
zEC12
z196
Model # PUs
Max
Memory
NE1 141 10 TB
NC9 129 10 TB
N96 96 7.5 TB
N63 63 5 TB
N30 30 2.5 TB
Investment protection strategy supported by:
Announced upgrade paths from z196 and zEC12
N-2 Sysplex co-existence (backwards and forwards)
Ensemble co-existence
Node based zBX mod 004 eliminates tight coupling to CEC
12
2
63. 12/10/2015
63
Supported Operating Systems
Operation System Supported Levels
z/OS • z/OS V2.1 with PTFs (exploitation)
• z/OS V1R13 with PTFs (coexistance)
• z/OS V1R12 with PTFs (Toleration. Life cycle extension required)
z/VM • z/VM 6.3 (exploitation)
• z/VM 6.2 (toleration)
z/VSE • z/VSE 5.1 and 5.2 will support z13 GA1 in compatibility mode with PTFs
• z/VSE 5.2 (with PTFs) will exploit Crypto Express5s
zTPF • z/TPF 1.1
Linux1 • SLES 11 and SLES 12
• RHEL 6 and RHEL 7
1The intention is to support n and n-1 Linux distribution releases depending on how their product cycle rolls out. The
majority of z13 exploitation will be made available with the major distribution releases following System zNext general
availability. IBM intends to make selected exploitation items available on selected distributions and work with its Linux
distribution partners to make these available
A Complete Workload-Optimized System
Integration of operations and business-critical analytics into one streamlined, end-to-
end data lifecycle
126
Transact
Transform
Analyze
Report
Better business response
Reduced data movement, reduced complexity, reduced configuration resources
More accurate, more secure, more available
CUSTOMER
INTERACTION
DATA IN
BUSINESS
INSIGHTS
OUT
64. 12/10/2015
64
Sources of Information
IBM DB2 Analytics Accelerator
– Primary Product Page
– Prerequisites and Maintenance
– Guides and manuals
– Knowledge Center
Customer Testimonials
– https://engage.vevent.com/index.jsp?eid=556&seid=68284&code=brand
Redbooks and Redpapers
– Reliability and Performance with IBM DB2 Analytics Accelerator Version 4.1
– Optimizing DB2 Queries with IBM DB2 Analytics Accelerator for z/OS
– Hybrid Analytics Solutions using IBM DB2 Analytics Accelerator for z/OS V3.1
– IBM DB2 Analytics Accelerator: High Availability and Disaster Recovery
– SAP Integration with IBM DB2 Analytics Accelerator for z/OS
All TechDocs available at the following link.
127
65. 12/10/2015
65
Eight major trends that will affect the industry in
coming years
Business
Services
Hardware
Software
Technology
Services
Security
Through 2016, the financial
impact of cybercrime will
grow 10% per year, due to
the continuing discovery of
new vulnerabilities
Growth Markets
By 2015, IDC expects emerging
markets to generate over 33% of
all IT spending
Big Data
Through 2015, more than 85% of Fortune
500 organizations will fail to effectively
exploit big data for competitive advantage
Analytics
Through 2015, more than 90%
of business leaders contend
information is a strategic asset,
yet fewer than 10% will quantify
its economic value
Social Business
By 2014, 20% of business
users will replace email as
the primary interpersonal
communications with
social networking
Smarter Planet
Over $100 billion: Global
investment in technology to support
smart city development by 2020
Mobile Enterprise
66% of CIOs ranked mobility as a
top investment priority in 2012
Cloud
Economic benefits of cloud will
continue to be the #1 driver of
adoption through 2016 for most
companies.*
Strategic Market Trends
Spring 2012
Customer
Sets
*Source: IDC, IDC's CloudTrack 2012 Summer Survey, Part 1: Cost Savings in the Cloud, November 14, 2012
Business Initiatives
% 10% 20% 30% 40% 50% 60%
130
Q. In 2015, which of the following business initiatives will be significant in driving
IT investments at your organization?
N=242
Increase Productivity
Reduce Costs
Improve Business Processes
Increase Revenue
Introduce new Products & Services
Increase Agility
Improve Customer Retention
Source: IDC IT Experience Survey, January 2015
66. 12/10/2015
66
Power & Cooling
Mgmt & Administration
New Server Spending
Why OPEX Matters
IT Efficiency? IT Effectiveness?
131
1995 2000
2005
2010
2015
OPEX/CAPEX Ratio
0.48 0.77
1.51
2.83
3.91
$94B
$123B
$138B
$206B
$272B
WW Server, P&C and
Administration Spending
3x
8x
2009
800,000 petabytes
2020
35 zettabytes
as much Data and Content
Over Coming Decade
44x Business leaders frequently
make decisions based on
information they don’t trust, or
don’t have
1in3
83%
of CIOs cited “Business
intelligence and analytics” as
part of their visionary plans
to enhance competitiveness
Business leaders say they don’’’’t
have access to the information
they need to do their jobs
1in2
of CEOs need to do a better job
capturing and understanding
information rapidly in order to
make swift business decisions
60%
Of world’s data
is unstructured
80%
Big Data Big Value
to enterprise and society
The resulting explosion of information (plus intermediate data) creates a need for a new kind of intelligence
Kilobyte (kB) 1,000 Bytes
Megabyte (MB) 1,000 Kilobytes
Gigabyte (GB) 1,000 Megabytes
Terabyte (TB) 1,000 Gigabytes
Petabyte (PB) 1,000 Terabytes
Exabyte (EB) 1,000 Petabytes
Zettabyte (ZB) 1,000 Exabytes
67. 12/10/2015
67
Conclusion
• There is no shortage of Big Problems that require Big Data
• The Nature of Data in IT is changing.
• Volume – Data doubling every two years
• Variety – Heading to a trillion devices; Unstructured data;
• Velocity – Sometimes all you have is milliseconds to respond
• Veracity – My business, finances, safety, health, life depend on
• Not all Big Data systems are created equal even if the datasheet says they are!
13
4
Big
Data
Big
Data
Big Data allows to bring together all kinds
of data on one single platform
68. 12/10/2015
68
135
Unstructured Data
Documents
(Word, Excel, PDF, TXT, etc.)
Social Media Data
Facebook, Twitter,
YouTube, Internet, etc.
Multimedia Data
Films, Music, Pictures, etc.
Events / Streams
Live-Cam, Exchange-Trigger,
Microphone, etc.
Operational Data
Policy, Claim, Underwriting,
General Ledger, CRM, etc.
Big
Data
on z
Big Data allows to bring together all kinds
of data on one single z13 platform
Most of the structured data used today resides there already, and extracting golden nuggets from
unstructructured data has never been easier
136
Big Data Analytics allows various kinds
of Analytics on one single z13 platform
Real time, predictive and in-transaction. Bring analytics to the data.
Analytics on Databases
BigSheets, reports, dashboards, etc.
on unstructured data
Operational Analytics
Reports, dashbords, etc.
on operational data
(i.e. legacy systems)
Business Analytics
Reports, dashbords, etc.
on dispositive Data
(i.e. Data Warehouse)
Predictive Analysis
What-if-Analysis,
Cluster-Analysis, etc.
Visualization/Discovery
Search, connection and
visualization of data of different
datasources and –types with one
application
Realtime Analytics
Dashboards, reports of events and
streamed data
69. 12/10/2015
69
Top 5 Use Cases for z13
Risk, compliance and counter fraud detection
z13 can deliver real time analytics and scale up to meet future demand
Operational Insight
z13 can deliver customer insight and next-best-action with a competitive TCO
Payment
z13 is the industry-proven platform for realtime payment and transaction
management
Cloud, API management and Linux
z13 can deliver the required scaleability for private clouds, API management
platforms and critical Linux-based workload
Mobile
z13 can scale up to meet the high-volume transactions demand from mobile
YouTube z13 Nordic Videos
13
8
https://www.youtube.com/watch?v=YuytuCEYIUg
https://www.youtube.com/watch?v=sl8VKXiatjM
https://www.youtube.com/watch?v=Bi4-0guBbOI
70. 12/10/2015
70
How many times have you used a Mainframe today?
IBM Mainframe: Make the Extraordinary Possible
http://www.youtube.com/watch?v=0tgt4VSrPso
Mainframe 50
IBM Mainframe: Make the Extraordinary Possible
http://www.youtube.com/watch?v=x6MpJL9XBlU
71. 12/10/2015
71
Server Consolidation on System z
http://www.youtube.com/watch?v=mMsvTR0x334
z13: Installations
Status per October 2015
500+ thus far worldwide (15+ Nordic)
– Most installations, migrations and upgrades
went smooth w/o problems
– What we see:
• Most installation are within -1% to +8% of the
zPCR modelled expectation
• Expectation is ~10% better ITR than zEC12
– Plan upgrade carefully
• So far few installations experience some
problems with the migration:
– Less compared to previous significant
processor design changes (for example
z10)
– But enough to warrant planning
considerations
72. 12/10/2015
72
Matthias R. Bangert, Executive IT-
Specialist
IBM Europe, z Systems
November 14, 2015
Exploiting IBM z13 to the utmost
(these results were NOT measured in an “IBM controlled environment” – it is CUSTOMER
PRODUCTION data)
• Java 7 to Java 8 migration: - 25% CPU consumption
• DB2 V10 to V11 migration: - 15% CPU consumption
• SIMD usage in Assembler programs: - 30% CPU consumption
• Cobol 4 to 5.2 migration (project started): - 20% CPU consumption (expected)
• MQ Series V7 to V8 migration: up to 50% less batch elapsed time
• SMT exploitation for ZIIPs: + 20% capacity
• Large memory exploitation: VSAM/RLS, DB2 Buffer Pools,
MQ Series V8
The benefits this particular client got out of their z13 clearly shows, that a holistic approach to gain the
most out of z13 is not optional – it is mandatory. It also confirms the results the z13 Benefit Estimator
tool is calculating.
IBM LinuxONE In Action…
IBM LinuxOne In Action: Scalable Financial Trading
https://www.youtube.com/watch?v=VWBNoIwGEjo
“I demonstrate the new IBM LinuxONE system for scalable
financial trading at the LinuxCon 2015 conference. The demo
shows multiple data loads (live data from the S&P 500 and
Tweets) streaming via Maria DB, MongoDB, Spark Analytics,
Chef, Docker and PostgreSQL.
In this LinuxONE demo, even with drastic upticks in CPU
Utilization during the Greek financial crisis, response times are
still lightning fast”.
74. 12/10/2015
74
z13 Planning: What else can be done?
• Plan your z13 LPARs carefully
– Why is this important?
• We will see that the LPAR layout matters
– That means LPAR weight, # of logical processors and the resulting numbers of
vertical high, medium and low processors (VH, VM, VL)
• A good starting point here is the LPAR Design tool available from the WLM homepage:
http://www-03.ibm.com/systems/z/os/zos/features/wlm/WLM_Further_Info_Tools.html#Design
• What does it do?
– It allows you to layout your LPARs
– Examine the VH, VM, VL processors
– Provides guidelines on how to set up the LPARs efficiently
Preparation for z13: CPU MF
• Use the CPU Measurement Facility (Hardware Instrumentation
Sampling) to obtain insight into the processor and cache architecture
• Value of CPU Measurement Facility (CPU MF)
– Recommended methodology for successful z Systems processor
capacity planning
• Need on “Before” processor to determine LSPR workload
– Validate achieved z Systems processor performance
• Needed on “Before” and “After” processors
– Provide insights for workload pattern, behavior, new features and
functions
• Continuously running on all LPARs
Capturing CPU MF data is an industry “Best Practice”
75. 12/10/2015
75
Preparation for z13: CPU MF …
• Introduced in z10 and later processors
• Facility that provides hardware instrumentation data for production systems
• Two major components
– Counters – for capacity planning
• Cache and memory hierarchy information
• SCPs supported include z/OS and zVM
– Sampling – for detailed, module level analysis
• z/OS HIS started task
– Gathered on an LPAR basis
– Writes SMF 113 records
• z/VM Monitor Records
– Gathered on an LPAR basis – all guests are aggregated
– Writes new Domain 5 (Processor) Records 13 (CPU MF Counters) records
• Minimal Overhead
A plug for
Cheryl Watson’s
Tuning Letter:
User experiences...
Frank’s Viewpoint...
zIIP Capacity Planning...
Prep for VSAM/RLS...
z13 Performance...
Software pricing workshops
150
76. 12/10/2015
76
Cheryl Watson Tuning Letter
article on z13 performance and related items – 2015 no. 2
151
Large Systems Update (LSU) 2015
• Stockholm 9-10 Nov
• Oslo 11-12 Nov
• Helsinki 16-17 Nov
• Copenhagen 18-19 Nov
• http://www-03.ibm.com/systems/no/lsu2015/index.html
152