Seamless FPGA deployment over Spark in cloud computing: A use case on machine learning hardware acceleration

Christoforos Kachris, Elias Koromilas, Ioannis Stamelos, Dimitrios Soudris
kachris@microlab.ntua.gr
ICCS-National Technical University of Athens
ARC 2018, Santorini
Seamless FPGA deployment over Spark in cloud
computing: A use case on machine learning
hardware acceleration

www.vineyard-h2020.eu
Network traffic in the data centers
2
Christoforos Kachris, ICCS, Greece

Power consumption in the data centers
3
• Currently Data
Centers consume
huge amounts of
energy
• Servers consume
around 30% of the
total power budget of
the IT infrastructure

Diverse Data Center Demands
4

FPGAs at the spotlight
5
April 2015
Submission of VINEYARD proposal
2015
2016
2017
December 2016
Overall, Intel now has five different AI
platforms; FPGAs, the Xeon Phi, the
Nervana NNP, the Myriad X, and its
traditional Core processor. The Core
processor still performs most AI tasks.

FPGAs in the news
Christoforos Kachris, ICCS, Greece 6

FPGAs in Data Center
• Intel: “Two orders of magnitude faster than GPU by 2020”
($16.7 billion bet)
Broadwel Xeon with Arria 10
• Microsoft Bing with Altera Stratix V
• IBM SupperVessel with Power8 + Xilinx
• Xilinx SDAccel on Nimbix Cloud
• Google has released TPU only for Tensorflow – ISCA 2017
7

Machine learning market size
• The machine learning market
size is expected to grow from
USD 1.41 Billion in 2017 to
USD 8.81 Billion by 2022, at a
Compound Annual Growth
Rate (CAGR) of 44.1%.
https://www.marketsandmarkets.com/PressReleas
es/machine-learning.asp

Machine learning as a service

Apache Spark
The largest open source project in
data processing.
• Structured Data
• Streaming Analytics
• Machine Learning
• Graph Computation
Provides an interface for
programming entire clusters with
implicit data parallelism and fault-
tolerance.
10

Contributions
• The FPGA driver API is packed in a shared object library and
can be used in a transparent way hiding all the low level
details.
• We implemented top level APIs in Python for standalone and
Apache Spark integrated use, that are easy to be used and are
also easily maintained since the middle layer, our shared
library remains the same for all of the above.

System stack
• Application Layer: This layer
hosts users’ applications. The
applications can run natively
using Python.
• Vineyard Layer: This layer hosts
the whole functionality of our
framework. The key element of
this layer is the implemented
shared library
• SDSoC-HLS API and FPGA
layerhared library

Flow for data movement - RDDS
• Flow of the original
and optimized
method for the
DMA transfers to
the accelerator

Use case: K-means clustering

Evaluation platforms

Pynq: Python Productivity for Zynq
• An open-source project from Xilinx that
makes it easy to design embedded
systems with Zynq MPSoCs.
• The APSoC is programmed using
Python.
• The code is developed and tested
directly on the PYNQ-Z1 board.
• The programmable logic circuits are
imported as hardware libraries and
programmed through their APIs in
essentially the same way as the
software libraries.
16

Spark integration on Heterogeneous MPSoC
17
K-means clustering

Spynq: Spark on Pynq integration
18

Speedup

Energy consumption
• Intel Xeon vs Pynq
cluster
• ARM vs Pynq cluster

Available on github
21

Cluster of Zynq (Pynq devices) running Spark
22

SW on Intel Xeon
23
Accel on FPGA MPSoC

VINEYARD Framework
24
• Accelerators stored
in an AppStore
• Cloud users request
accelerators based
on applications
requirements
• Decouple Hardware
– Software
designers
Cloud computing Applications
VINEYARD Cloud Resource Manager
3rd party IP
developersLibrary of Hardware
accelerators as IP
Blocks
Heterogeneous Data Center
DFE
Processors Dataflow Proc.+FPGA
IP Accelerator’s
App store
Cloud tenants
Acc
Acc
Acc
Acc
DFE
DFE
DFE
Accelerator Controller
Accelerator Virtualization
Scheduler
Accelerator API
Performance
Energy

Consortium
26
Platform Evaluator
Data Centre Vendor
System Vendor (Dataflow Eng.)
System Software
Programming framework &
Hardware accelerators
Data Centre
Software developers
Data Centre End User

VINEYARD appears in Xilinx’s website
27

Ecosystem on FPGAs in the cloud

Main goals
VINEYARD AIMS TO
• Build an integrated platform for energy-efficient data
centres based on novel programmable hardware
accelerators
• 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 data-centre programming
frameworks (i.e. Spark.).
• VINEYARD also foster the establishment of an
ecosystem that will empower open innovation based on
hardware accelerators as data-centre plugins for
marketplace, thereby facilitating innovative enterprises
(large industries, SMEs, and creative start-ups) to
develop novel solutions using VINEYARDS’s leading
edge developments.
29

• Speedup your application seamlessly
• An integrated framework for the utilization of hardware
accelerators in HPC and data center seamlessly
Contact detais: kachris@microlab.ntua.gr

Seamless FPGA deployment over Spark in cloud computing: A use case on machine learning hardware acceleration

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Seamless FPGA deployment over Spark in cloud computing: A use case on machine learning hardware acceleration

Similar to Seamless FPGA deployment over Spark in cloud computing: A use case on machine learning hardware acceleration (20)

Recently uploaded

Recently uploaded (20)

Seamless FPGA deployment over Spark in cloud computing: A use case on machine learning hardware acceleration