This document discusses using Docker containers on OpenPOWER systems for machine learning applications. Some key points: - OpenPOWER systems have advantages for machine learning like more CPU cores, threads, memory and I/O bandwidth which allow scaling out training across systems. - Docker allows distributing machine learning models easily across OpenPOWER systems for faster training times. - The same Docker containers can run on x86 and OpenPOWER systems, providing a consistent developer experience for machine learning applications. - OpenPOWER supports technologies like GPUs and FPGAs which provide huge speed-ups for machine learning and deep learning analytics.

1. Build FAST Learning
Apps with Docker and
OpenPOWER
Indrajit Poddar (I.P), Seetharami Seelam
IBM

2. IBM Systems
Deep Learning
What you and I (our brains) do without even thinking about it…..we recognize a bicycle

3. IBM Systems
Now machines are learning the way we learn….
3
From "Texture of the Nervous
System of Man and the Vertebrates"
by Santiago Ramón y Cajal.
Artificial Neural Networks

4. IBM Systems
But training needs a lot computational resources
Easy scale-out with: Deep Learning model training is hard to distribute
Training can take hours, days or weeks
Input data and model sizes are becoming
larger than ever (e.g. video input, billions of
features etc.)
Real-time analytics with:
Unprecedented demand for offloaded computation,
accelerators, and higher memory bandwidth systems
Resulting in….
Moore’s law is dying

5. IBM Systems
OpenPOWER: Open Hardware for High Performance
5
Systems designed for
big data analytics
and superior cloud economics
Upto:
12 cores per cpu
96 hardware threads per cpu
1 TB RAM
7.6Tb/s combined I/O Bandwidth
GPUs and FPGAs coming…
OpenPOWER
Traditional
Intel x86
http://www.softlayer.com/bare-metal-search?processorModel[]=9

6. IBM SystemsDemo Shown at DockerCon Europe 2015
Total: 10,011 Containers on One System: Ubuntu(8028), Node.js(991), Wordpress(992)
• Wider, Faster Memory Interface, Faster Cores with More Threads
• Split-Core Mode supports Interactive Web style Apps better
• 2x Greater Density of Containers per systems lowers Cost
• >40% better Throughput and 4x better Latency
• OpenPower ecosystem offers wide range of Open HW Platforms
Open Source Docker
Docker Containers running on Power have Superior Density
6

7. IBM Systems
A Consistent Developer Experience
7
docker pull ubuntu:latest
will get you the POWER/LinuxOne/X86 specific ubuntu image!!!
Base Image
X86
Node.js App
X86 Node.js
runtme
Base Image
Power
Node.js App
Power Node.js
runtime
Base Image
Z
Node.js App
Z Node.js
runtime
Docker on
X86
Docker on
Power
Docker on
Z
Node.js App2 Node.js App2 Node.js App2
Container Container Container
Multi-platform Docker images

8. IBM Systems
OpenPOWER: GPU support
8
Credit: Kevin Klaues, Mesosphere
IBM Spectrum
Conductor includes
enhanced support for
fine grained GPU and
CPU scheduling with
Apache Spark and
Docker
Mesos supports GPUs
Huge speed-ups with GPUs and OpenPOWER!

9. IBM Systems
Machine Learning and Deep Learning analytics on OpenPOWER
No code changes needed!!
9
ATLAS
Automatically Tuned Linear Algebra
Software)

10. IBM Systems
Learn More and Get Started…
10
Power-Efficient Machine Learning on
POWER Systems using FPGA Acceleration
Machine and Deep Learning on Power Systems
Register for a SuperVessel Account and take deep learning
notebooks running in docker containers a spin!
https://ny1.ptopenlab.com/bigdata_cluster