Christine McMonigal, Storage SW Ecosystem Manager, Data Center Group, Intel

1. Christine McMonigal, Storage SW Ecosystem Manager, Data Center Group
1

2. DCG Storage Group 2
Driven by:
 Mobile
 Social Media
 Internet of Things
 Big Data and Cloud
Source: IDC – The Digital Universe of Opportunities: Rich Data and the Increasing Value of the Internet of Things - April 2014
StorageCostStructureNeedsaFundamentalShift
1 2 3 5 7
12
19
30
48
77
125
2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023
Storage Capacity
in Terabytes
62%CAGR
2%
IT Budget
CAGR
As data volumes continue to grow, IT needs storage solutions that can reduce costs
while still meeting other storage requirements

3. 3
STORAGE
TRANSFORMING
KEY DRIVERS
CLOUD
CLOUD SERVICES
1
STORAGE MEDIA
TRANSITIONS
3 3D XPOINT™
TECHNOLOGY
NEXT GENERATION
ARCHITECTURES
2
CLOUD
CLOUD SERVICES
NEXT GENERATION
ARCHITECTURES
STORAGE MEDIA
TRANSITIONS
3D XPOINT™
TECHNOLOGY

4. DCG Storage Group 4
Today’sInfrastructure
storage storage
Cloud
STORAGE
Enterprise
STORAGE
Themissinglink
Ability to seamlessly access data
anywhere, at anytime, on any device
at the required performance

5. DCG Storage Group 5
WHYSTORAGEMODERNIZATIONISESSENTIAL
Internet of Things
Media Transition Scale Out
Scale Up
Storage Modernization: Seamless data access anywhere,
at anytime, on any device at the required performance
AGILE, Automated & Secure Infrastructures
and Business Models
Cloud
Enterprise
01
1010101
101010101010
010101010101
101010101010
Orchestration

6. DCG Storage Group
MovefromScale-uptoScale-OutArchitectures
6
Supports storage capacity growth cost effectively
Standard Ethernet
network
Data distributed across
multiple nodes or clusters
Flexible design to support
multiple workloads
Separate,
dedicated networks
Data stored in proprietary
storage hardware
Optimized to run only a
specific workload
Compute nodes working together
external network
Scalesperformance&capacity
Scale-OutScale-up
Single System
internal network
SCALESBYADDINGDISKSFORCAPACITY

7. 7
openstack(andopensource)scale-outstorageBusiness Drivers from OpenStack User Survey April 2016

8. DCG Storage Group
OpenSourceScale-OutCephwithoptimalcommercialsupport
8
Clients
Ceph Storage Cluster
Server 1
Ceph
OS
CPU
Memory
Disks
Network
Server 2
Ceph
OS
CPU
Memory
Disks
Network
Server 3
Ceph
OS
CPU
Memory
Disks
Network
Server 4
Ceph
OS
CPU
Memory
Disks
Network
Object Storage Block Storage File Storage
One platform that provides
Object Storage, Block
Storage, S3 interface and File

9. StorageEvolution
9Technology claims are based on comparisons of latency, density and write cycling metrics amongst memory technologies recorded on
published specifications of in-market memory products against internal Intel specifications.

10. 10
Intel’sroleinstorage
AdvancetheIndustry
OpenSource&Standards
BuildanOpenEcosystem
Intel®StorageBuilders
Endusersolutions
Cloud,Enterprise
IntelTechnologyLeadership
Storage Optimized Platforms
Intel® Xeon® E5-2600 v4 Platform
Intel® Xeon® Processor D-1500 Platform
Intel® Converged Network Adapters 10/40GbE
Intel® SSDs for DC & Cloud
Storage Optimized Software
Intel® Intelligent Storage Acceleration Library
Storage Performance Development Kit
Intel® Cache Acceleration Software
SSD & Non-Volatile Memory
Interfaces: SATA , NVMe PCIe,
Form Factors: 2.5”, M.2, U.2, PCIe AIC
New Technologies: 3D NAND, Intel® Optane™
Ceph community contributions on
workload profiling, latency analysis
and performance optimizations
90+ partners
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific
computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you
in fully evaluating your contemplated purchases, including the performance of that product when combined with other products.
Storage solutions architectures
Intel solution architects have deep
expertise on Ceph for low cost
and high performance usage
helpingcustomerstoenableamodernstorageinfrastructure

11. How well can
Ceph perform?
Which of my workloads
can it handle?
Which server
hardware is required
for it to perform well?

12. Optimization
Criteria
Properties Example Uses
IOPS-optimized  Lowest cost per IOPS
 Highest IOPS
 Meets minimum fault domain recommend-
ation (single server is less than or equal to
10% of the cluster)
 Typically block storage
 3x replication on hard disk
drive (HDD) or 2x replication
on Intel® SSD DC Series
 MySQL on OpenStack clouds
Throughput-
optimized
 Lowest cost per given unit of throughput
 Highest throughput
 Highest throughput per BTU
 Highest throughput per watt
 Meets minimum fault domain recommend-
ation (single server is less than or equal to
10% of the cluster)
 Block or object storage
 3x replication
 Active performance storage
for video, audio, and images
 Streaming media
Capacity-
optimized
 Lowest cost per TB
 Lowest BTU per TB
 Lowest watt per TB
 Meets minimum fault domain recommend-
ation (single server is less than or equal to
15% of the cluster)
 Typically object storage
 Erasure coding common for
maximizing usable capacity
 Object archive
 Video, audio, and image
object archive repositories

13. Open Stack Starter S M L
64TB 256TB + 1PB + 2PB +
IOPS
Optimized
• Ceph block (RBD)
• Intel® P3700s w/ co-located write journals
OR
• Intel® S3610 w/ P3700 write journals in a 4:1 ratio
• Multiple OSDs per flash drive
• 10 Intel® Xeon® E5 cores per P3700; 4 per S3610
• 2x or 3x replication (with backup)
Throughput
Optimized
• Ceph block or object (RBD or RGW)
• HDDs w/ P3700 or S3710 write journals
• 1 Xeon core per 2 HDDs (e.g. with 24 HDDs, 12 core Intel® Xeon® E5-2650 v4)
• Single OSD per HDD
• 10GbE  40GbE with >12 HDD per chassis
• 3x replication
Cost-Capacity
Optimized
• Ceph object (RGW)
• HDD drives with no SSD journal
• 1 Xeon core per 2 HDD
• Single OSD per HDD
• Erasure-coded

14. 14
Recent Intel testing shows it’s
possible to reach 1.4M IOPS
Cephfordatabaseworkloads
Using a MySQL database
workload, we measured 400K
OLTP QPS using a mix of
70/30% Select/Update

15. DCG Storage Group
Cephperformancerange
15
Perceived performance range
Actual performance range
(measured)

16. DCG Storage Group
Storagemodernization
16
½ latency
2x throughput
50% CapEX reduction
Storage
Intel® Differentiated
Storage ServicesApps
Intel® CAS
“Web”
“Analytics’”
..
.
1IDF 2015 Class SDS002 http://intelstudios.edgesuite.net//idf/2015/sf/aep/SSDS002/SSDS002.html
IDF2015 Technology Demo: https://www.youtube.com/watch?v=vtIIbxO4Zlk
Results of Yahoo* internal benchmark testing - Ruiping Sun, Principal Architect, Yahoo*. Hardware/Software Config: 8 OSD Nodes, each: HP ProLiant DL180 G6 ySPEC 39.5,
2x Xeon X5650 2.67GHz (HT enabled, total 12 cores, 24 threads), Intel 5520 IOH-36D B3 (Tylersburg), 48GB 1333MHz DDR3 (12x4GB PC3-10600 Samsung DDR3-1333
ECC Registered CL9 2Rx4), 10*8TB 7200 RPM SATA HDDs, 1*1.6TB Intel P3600 SSD (10GB journal per OSD, 1.5TB cache) (CAS config only), 2*HP NC362i/Intel 82576
Gigabit NICs, 2*Intel 82599EB 10GbE NICs, RHEL 6.5 w/kernel 3.10.0-123.4.4.el7

17. DCG Storage Group
Availableresources
17
https://www.redhat.com/en/resources/
red-hat-ceph-storage-servers-intel®-
processors-and-ssds
https://www.percona.com/resources
/videos/mysql-cloud-head-head-
performance-lab
https://rh2016.smarteventscloud.
com/connect/sessionDetail.ww?S
ESSION_ID=88853&tclass=popup
Watch for it posted to
www.intel.com by Friday!

18. SummaryandNextSteps
 Ceph can help deliver on the promise of the cloud, using next
generation storage architectures
 Flash technology enables new capabilities in small footprints
 Ceph and MySQL provide a compelling case for converged
storage that can support latency sensitive analytics workloads
 Next steps:
 Access the available resources to learn more about how Ceph can be
deployed for your workloads
 Consider a Ceph pilot with Red Hat and Intel
Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Any difference in system hardware or software design or configuration may affect actual performance. See
configuration slides in backup for details on software configuration and test benchmark parameters.
18

20. DCG Storage Group
WhatWorkloadsareDrivingPublicCloudStorage?
20
Consumer-generated content creates greater
demand for:
 Media Streaming
 Cloud Storage
 Social Media
Enterprise requirements create demand for:
 Database
 Online Backup / Disaster Recovery
 Big Data / Analytics
 Media

21. Pushing Buying Behaviors Away from Norm
Cost efficiency
Speed of provisioning
Vendor lock-in
Storage scaling
Admin
Governance
IOPS per GB
better
faster
less lock-in
elastic pool
more expertise
more control
broad options
better
faster
more lock-in
silos*
less expertise
less control
narrow options
public cloud
storage
NAS/SAN
storage appliances
software defined storage
commodity servers

22. Illustrating 10x Range of Ceph Performance
1 DVD movie per second with 3-node cluster
(standard 2U Ceph servers, hybrid)
E-Book
MP3 Song
Audio CD
DVD Movie
HD Movie
Blue-ray Movie
UHD Movie
1 Blue-ray movie per second with 3-node cluster
(standard 2U Ceph servers, all-flash)

23. Range of Ceph Use-Case Examples
Workload
IO Profiles
Workload
Examples
Searchable
Examples
Workload IO
Characteristics
Hardware
Characteristics
IOPS MySQL
MariaDB
PostgreSQL
Medallia High IOPS/GB
Smaller random IO
Read/write mix
Sled-based chassis
Intel® DC P3700
10 Xeon® cores per P3700
10GbE
Balanced/Throughpu
t
Digital media distro
Server virtualization
(OpenStack Cinder)
Bloomberg
Target, Walmart
yahoo!, Comcast
High MBps/TB
Larger sequential IO
Read/write mix
Standard -> dense chassis
HDD -> Intel® DC P3700
Balanced core : drive ratio
10GbE -> 40GbE
Capacity-Archive Digital media archive
Object archive
Big Data archive
yahoo!
CERN
Low cost/GB
Sequential IO
Write mostly
Dense -> ultra-dense chassis
HDD
Low core : drive ratio
10GbE