OpenPOWER hardware into a heterogeneous infrastructure Jobs Run on Mix Architecture While Users Get Coffee Creating a mix architecture that incorporates the best new hardware without the users knowing ID: 674131
Download Presentation The PPT/PDF document "The Palmolive Effect: A model for implem..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
The Palmolive Effect: A model for implementing OpenPOWER hardware into a heterogeneous infrastructure Slide2Slide3
Jobs Run on Mix Architecture While Users Get Coffee
Creating a mix architecture that incorporates the best new hardware without the users knowing
allows for deeper processing of data. Users inherently no nothing about architectures or even what they are currently running on in terms of desktop computers. Why would we as administrators force them to manage workloads across machines for which they have no formal knowledge. It is our responsibility to lower the activation energy for all users to take advantage of resources without having to gain specific knowledge.
Slide4
What is the CGRB at Oregon StateWhy Do Computational Methods and Techniques Need a Heterogeneous Environment
How to Manage User Setting Across Multiple ArchitecturesScheduler’s Help Users Run Job Across Different Architectures.
Running Jobs within a Heterogeneous Architecture.
Examples of building tools on
OpenPOWER
based systems. Slide5
Core Facilities & Computational
Science
Building infrastructure for Researchers
Tools for Data
Mining
& Data Processing
Building
New Algorithms /
N
ew Tools
Creating Deliverables for PublicationsWays to Reduce Cost and Increase ScopeSlide6
CGRB Computational Infrastructure
Started out as a single machine in 2000 (4 processors, 4G RAM)
A small cluster was added in 2002 allowing users to process more then a limited number of jobs at a time (20 machine, 40 processor, 512G RAM). The CGRB grew this over time to always provide a limited set of resources to all users.
Groups were encouraged to add to the cluster
with processing hardware included into grants to support research work.
Researchers pay for having hardware managed by the core facility
and included into our infrastructure with access provided to their research group.
CGRB created a storage server
where users can add storage to support research at a cost model that can be included into grants.
CGRB created pathways allowing general users to
rent
processing
, rent storage, web and time from our core facility.
Building InfrastructureSlide7
GENOME Cloud
~20,000 Jobs / Day
4100
Processors
3.5+
PB Redundant
Storage
10 machines with greater then 1TB of memory
6x POWER8 Systems
Increase access to resources
Decrease analysis time
Increase Network Speed
CGRB InfrastructureSlide8
Why Do Computational Methods and Techniques Need a Heterogeneous Environment Slide9
Limits Users and Researchers to one set of tools.
No ability to negotiate price since your stuck in a single architecture.
Many x86 based machines get easily overloaded.
Limits to
Input/Output
pathways.
Single processor vendor put the users in line with only that forward pathway.
Non-
Heterogenous
InfrastructureSlide10
Provides
u
sers with a cafeteria of tools coming from any architecture.
Can
negotiate price since users can use any architecture
that can run the tools.
Allows users to find pathways around
l
imits on
Input/Output
pathways.Multiple processor vendors put the users into the forward pathway that best fits their needs. Heterogeneous InfrastructureSlide11
How to Manage User Setting Across Multiple ArchitecturesSlide12
Use shell and environmental settings to help manage user information across multiple architectures.
Things like “
uname
” will provide information about platform and other important configurations.
Use different global settings files that users can source to provide architecture specific paths and settings.
Separate directories for all binaries of a different architectures so users can easily work with tools.
Use Environment SettingSlide13
CSH
BASH
Using “
uname
” to set ARCHSlide14
Using “
uname
” to set ARCH (
csh
)Slide15
Using “
uname
” to set ARCH (
csh
)Slide16
Scheduler’s Help Users Run Job Across Different ArchitecturesSlide17
Use of schedulers will allow users/services to submit jobs to the infrastructure.
Legacy schedulers are more aware of a mixed architecture.
Using the environmental variables set to a specific architecture users can freely submit jobs knowing the system will find the correct binary for each system.
Scheduler will Help UsersSlide18
Gridware
was sold to and improved by Sun Microsystems and became known as Sun Grid Engine (SGE), CODINE (Computing in Distributed Networked Environments) or GRD (Global Resource Director
). This tool created a
grid computing computer cluster software system (otherwise known as a batch-queuing system
). There
have been open source versions and multiple commercial versions of this technology, initially from Sun, later from Oracle and then from
Univa Corporation.Grid Engine / SGE / OGE / Son of GE / UnivaSlide19
Interacting with Machines and Running JobsSlide20
Benefits of a Heterogeneous EnvironmentSlide21
Working with IBM and
OPENPower
the CGRB was able to test a new GZIP CAPI card
.
Massive increase of speed to compress and de-compress standard
gzip
files. Reduces load on CPU resources allowing them to be used for real processing.
Jobs that took over 60 hours are not taking less then 1 hour.
Blog Post:
A Better Way to Compress Big
DataGZIP CAPI CardSlide22
CGRB Uses Development EnvironmentNew Sequence Alignment Tool using GPU Hardware
With the CAPI and NVLink built into the OpenPOWER hardware we wanted to see if we can do real work on the hardware by means of genomic sequence alignment.
CASSA
-
C
UDA Accelerated Scalable Sequence AlignerCGRB has developed a new GPU based HTS alignment tools using the new IBM and NVIDIA hardware.CASSA can run the Bowtie2 or BWA seed methods.
CASSA runs everything on the GPU so it can run on Windows as well as POWER8 and x86 based Linux distributions. Slide23
CASSA PerformanceSlide24
CGRB IBM Collaboration Summary
Using IBM Power8 the CGRB was able to increase processing throughput 2-10x just by compiling software.
CGRB was able to work with IBM to continue porting software which was both beneficial to our users and IBM.
CGRB used the new development environment to create a new tool called CASSA to reduce time running HTS sequence alignment.
CASSA will be provided to the entire research community as soon as possible. Slide25
Summary
Computational Science is used in a gamut of research and clinical areas.
Many times new technologies and hardware change the way computational science can be done.
New technologies can generate magnitudes of order greater precision, change the scope of work or reduce bias.
New technologies many times require changes in tools and hardware.
Running
a heterogeneous
infrastructure will provide your group with the greatest flexibility to take on the future of computing.
Using different architectures in the same infrastructure is easy
...Slide26
#######################################################################
#########################
##
Acknowledgements
##
####################################################
##########################
##################
>
Software Tool
CentOS
Linux
Open Source Software (GNU
)
Son of Grid
Engine (SGE
)
NVIDIA
Cuda
>NVIDIA
Jon
Saposhnik
Robert
Crovella
>CGRB Staff
Ryan Kitchen
Shawn
O’neil
Ian Munoz
Brent
Kronmiller
Matthew
Peterson
>
Nimbix
Cloud Services
Leo Reiter
Tom McNeill
>IBM
Charles J.
Foretich
Keith Brown
Stan
Gowen
Terry
Leatherland
Denise
Ruffner
Indrajit
Poddar
Hal Porter
Linton Ward