VINEYARD Overview - ARC 2016

VINEYARD project:
Versatile Integrated framework for
Accelerator-based Heterogeneous
Data Centres
International Symposium of Applied Reconfigurable Computing, March 2016
Christoforos Kachris, Dimitrios Soudris
ICCS/NTUA, Greece
1

Current Data Centers
By 2018, more than three quarters (78%) of workloads will be processed
by cloud data centers.
[Source: Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update 2014–2019 White Paper]
2

Power budget has reached its limit
The power budget per processor has reached its limit. We can increase the number of
cores but we can no longer power all of the processors at the same time.
[Source: HiPEAC Vision 2015]
3

The data deluge gap
The Moore’s law cannot follow the increased growth of the data traffic that needs to be processed.
International Symposium of Applied Reconfigurable Computing, March 2016 4

Power consumption of Data Centers
• Currently Data
Centers
consume huge
amounts of
energy
• Servers
consume
around 30% of
the total power
budget

Hardware accelerators
• HW acceleration can be used to reduce significantly the
execution time and the energy consumption of several
applications (10x-100x)
6

FPGAs in the Cloud
• Altera’s acquisition by Intel
• Microsoft’s catapult for Bing search
• IBM open power – CAPI interface with FPGAs
7

Heterogeneous DCs for energy
efficiency
“The only way to differentiate server offerings is through accelerators, like we saw with cell
phones”, OpenServer Summit 2014 Leendert Van Doorn; AMD
TODAY’s DCs Future Heterogeneous DCs
with VINEYARD infrastructure
Run-time manager and
orchestrator
3rd party HW
accelerators
Run-time scheduler
Big Data Applications
• Low performance
• High power consumption
• Best effort
• Higher performance
• Lower power consumption
• Predictable performance
Requirements
Servers
P
Servers
P
P
P
P
P
P
P
P
P
P
P
DFE
DFE
DFE
DFE
VINEYARD
Servers with
dataflow-based
accelerators (DFE)
8

VINEYARD’s goals
VINEYARD aims to:
• Build an integrated platform for energy-efficient data
centres based on novel programmable hardware
accelerators (i.e. Dataflow engines and FPGA-coupled
servers).
• Develop a high-level programming framework and
big data infrastructure for allowing end-users to
seamlessly utilize these accelerators in
heterogeneous computing systems by employing
typical cloud programming frameworks (i.e. Spark).
The main goal is to increase significantly the performance
and the energy efficiency of the data centers

VINEYARD accelerators
VINEYARD will develop two types of hardware
accelerators:
• Dataflow engines: These accelerators will be
mainly used for applications that can be
represented mainly as data-flow graphs
• FPGA-based engines: These servers will be based
on MPSoC FPGA that incorporate multiple 64-bit
ARM cores and will be used for application that
needs low latency communication between the
processors and the accelerator

VINEYARD Heterogeneous
Accelerators-based Data centre
Bioinformatics Finance
VINEYARD Progr. Framework
Synthesis
(OpenSPL,OpenCL)
Pattern
Matching
Analytics
engines
String
matching
Other
processing
Commonly used
Function/tasks
…
HW Manager
Library of Hardware
functions as IP Blocks
Requirements:
• Throughput
• Latency
• Power
Racks with programmable
dataflow engine (DFE)
accelerators
Server Racks with
commodity processor
Repository
Compressi
on
EncryptionScheduler
DFE
DFE
DFE
DFE
Cluster Resource Manager
Analytics
P
P
P
P
P
P
P
P
Programm
able
Logic
Racks with
MPSoC FPGAs
Programming Framework, APIs
11

Objectives
• Objective 1: Development of novel Programmable
Dataflow Engines (DFE) for servers: One of the main
objectives of VINEYARD will be the development of novel
programmable dataflow engines (hardware accelerators)
based on coarse-grain programmable components that
can be coupled to servers’ processor in heterogeneous
data centres.
• Objective 2: Development of novel FPGA-accelerated
servers: VINEYARD will develop novel server blades that
will be based on high performance and energy-efficient
FPGAs that incorporate multiple low power cores.

Objectives
• Objective 3: Development of an open-source integrated
programming framework that can be used for the
programming of heterogeneous systems consisting of
general purpose processors (CPUs), and accelerators
(programmable dataflow engines and FPGAs) based on
traditional cloud programming frameworks (i.e. Spark).
• Objective 4: Development of a run-time
scheduler/orchestrator that controls the utilization of the
accelerators based on the applications’ requirements
(execution time, power consumption, available resources,
etc.).

Objectives
• Objective 5: Development of a novel Virtual-Machine
(VM) appliance model for provisioning of data to shared
accelerators. Targeting cloud deployments, this VINEYARD
effort will bring both tangible and novel results. The
enhanced VINEYARD middleware augments the
functionality of the orchestrator, by enabling more
informed allocation of tasks to accelerators.
• Objective 6: Ecosystem Establishment and Support. The
establishment of an ecosystem that will empower open
innovation based on hardware accelerators as data-centre
plugins, thereby facilitating innovative enterprises (large
industries, SMEs, and creative start-ups) to develop novel
solutions using VINEYARDS’s leading edge developments.

Overview of VINEYARD
Overall VINEYARD aspires to address the open challenges in integrating
programmable and hardware accelerators to the predominant
software stacks used for data analytics in the Cloud:
1. hide the accelerator from the programmer by presenting it as a
pure library function, embeddable in query processing, data
processing or aggregation tasks, and by extension to analytical
libraries written on top of high-level programming models;
2. extend the runtime systems of high-level analytics languages to
handle efficiently scheduling, communication, and synchronization
with programmable accelerators; and
3. improve the performance robustness of analytics written in high-
level languages against artefacts of virtualization, notably
performance interference due to contention on shared resources
and hidden noise in hypervisors and hosting VMs.

Consortium
Platform Evaluator
Data Centre Vendor
System Vendor (Dataflow Engines)
System Software
Programming framework &
Hardware accelerators
Data Centre
Software developers
Data Centre End User
16

The VINEYARD value-chain
VINEYARD framework
Soft IP-cores
vendor
Heterogeneous
Platform
Application
developers,
Cloud tenants
End user -
client
IP1
IP3
IP4
IP2
17

Three real-world scenarios
The VINEYARD project will be demonstrated on three real-word
applications:
• Computational neuroscience (Neurasmus):
• high-accuracy simulation of the Olivocerebellar system of the
brain, crucial to the understanding of brain functionality
• Financial applications (Neurocom Lux and ATHEX):
• Trading system operations
• Pre-trade risk management
• Data analytics (LeanXcale)
• TPC-C (on-line transaction processing (OLTP) benchmark)
• TPC-H (decision support benchmark).
• IoT application (Linear Road will also be used as a representative
workload in IoT applications)

VINEYARD details
• Project details
– Contract number: H2020- ICT 4 - 687628
– Community contribution: 6.28M€
– Start date: February 1st, 2016
– Duration: 36 Months
– Project Coordinator: Dimitrios Soudris, ICCS/NTUA,
dsoudris@microlab.ntua.gr
– Technical Project Manager: Christoforos Kachris, ICCS/NTUA,
kachris@microlab.ntua.gr
– Website: www.vineyard-h2020.eu

Thank you.
More information on www.vineyard-h2020.eu
Contact details:
Prof. Dimitrios Soudris: dsoudris@microlab.ntua.gr
Dr. Christoforos Kachris: kachris@microlab.ntua.gr
This project has received funding from the European Union’s Horizon 2020
research and innovation programme under grant agreement No 687628
20

VINEYARD Overview - ARC 2016

Recommandé

Recommandé

Contenu connexe

Tendances

Tendances (20)

Similaire à VINEYARD Overview - ARC 2016

Similaire à VINEYARD Overview - ARC 2016 (20)

Dernier

Dernier (20)

VINEYARD Overview - ARC 2016