Embed Size (px)
Citation preview
Katie AntypasNERSC User ServicesLawrence Berkeley National Lab
10 February 2012
JGI Compute User Training
Today we want to share plans, introduce new services, test workflows, answer questions and hear your feedback
New file systems and data management
Beta Web documentation
JGI User Survey Results
Fair share batch systems
Crius
RheaTheia
Kronos?
Hyperion
Oceanus
Iapetus
Themis
3
Breakdown of NERSC Users’ Science Areas
NERSC serves over 4000 users across 500 distinct projects across an array of science areas
JGI users are similar to traditional NERSC users in their need for:
JGI users have special workflow, throughput, and software needs
• Stable, reliable systems• Large data management and storage• Fast queue turn around• Access to millions of compute hours
5
Where we have come from …
One-on-one collaborations
MOU reached for NERSC to support JGI computational and IT systems
File system stabilization
Cluster Consolidation
2009
Spring 2010
Fall 2010-Fall 2011
May 2011- present
Crius
RheaTheia
Kronos?
Hyperion
Oceanus
Iapetus
Themis
Merge JGI systems into Crius
Crius
RheaThei
a
Kronos?
Hyperion
Oceanus
Iapetus
Themis
Next: Move Crius to NERSC Space
But … before we do this, we want to make sure all pipelines and workflows are tested so we cause minimum disruption to users
Primary benefit: Access to new 2PB file system
JGI Sys-ops members have been incorporated into NERSC groups
Ilya MalinovJeremy BrandMatt DunfordBrian YumaeRavi CheemaPatrick HajekFred Loebl
Networking, Security, Servers: Brent Draney
Continue to contact JGI sys-ops for day to day problems
Computational Systems Group: Jay Srinivasan
Storage Systems Group: Jason Hick
The IT Steering Committee makes policy decisions regarding the cluster
JGIAlex CopelandDaniel RokhsarHarris ShapiroHenrik NordbergIgor GrigorievJames BristowKostas MavrommatisLen PennacchioNikolaos KyrpidisRay TurnerRob EganVictor Markowitz
NERSCBrent DraneyJason HickJay SrinivasanJeff BroughtonKatie AntypasShane Canon
Contact your representative on the steering committee if you have concerns.
10
The NERSC consultants serve as user advocates
Woo-Sun YangTools/Math Libraries
Richard GerberAstro/Web Services
Helen HeClimate
Katie AntypasGroup Leader
Dave TurnerEverything
Zhengji ZhaoMat. Sci/Chemistry
Mike StewartCompilers
Harvey WassermanChemistry
??Bioinformatics
Consultant
Yushu YaoData Analytics
Jack DeslippeMat. Sci/Chemistry
Eric HjortHigh Energy
Physics
Logging into Phoebe
When you login to Phoebe you will be in your “global home” directory
• The full UNIX path is stored in the environment variable $HOME• Your $HOME quota is 40GB and 1,000,000 inodes
We realize most users have a different home directory in /house.
• Reference your old home directory as $OLD_HOME if you need • We use $HOME to initialize environment so do not redefine• Note /house is available on Phoebe, NetApps is not
Your default shell on Phoebe is bash
•NERSC sets .bashrc for all users as a read-only file.
•Do not change your .bashrc file!
•NERSC uses .bashrc file to make global configuration changes for all users
• Put your own customizations in .bashrc.ext
Want to change your shell? Just let us know.
When you login the environment is pre-configured for you
• /jgi/tools bin and lib directories in your path• Batch system environment setup
Jason HickStorage Systems GroupLawrence Berkeley National Lab
10 February 2012
A New 2PB GPFS file system for the JGI “projectb”
The new 2PB “projectb” file system is available on Phoebe now
• Some high level specs for users
• 2PB
• XXX
File systems best practices
• Unfortunately disk is still expensive
• All of the JGI’s data can not be stored on disk within the current budget
• Archive and delete data you no longer need
• Disk usage will be controlled through quotas in some cases and purging in others
There are two areas of storage within the “project” layout of the “projectb” file system
/projectb/
projectdirs/ scratch/
PI/ RD/ fungal/ metagenome/ micro/ plant/ user/
• Group directories• Not purged• Subject to quota
• User directories•Purged
It is important for every group to come up with a data retention policy
How long should we keep the raw data?
Can the data be deleted or should it be archived? Can we set up an
automated way to archive and delete data?
