User Group Executive Board Call

Present: Balint Joo, C.S. Chang, David Dixon, Katrin Heitmann, Mike Zingale, Joe Olfelein, Rangan Sukumar, John Turner, OLCF: TP Straatsma , Suzanne Parete-Koon, Adam Simpson, Veronica Larrea, Sherry Ray, Bill Renauld
Logistics:
Next Executive Board meeting will likely be held at SC14. Details will follow.

User Driven Content:

The motivation is to facilitate more interactions between OLCF projects, especially across fields, so they know who else is facing common problems.
Here are some ideas that we must review:

1. User Collaboration repositories where users could share open source code, such as that for an ODE solver optimized for hybrid architectures.
a. Code would be posted on a non-OLCF managed GitHub accounts. OLCF would provide a page of topics with links to these repositories.
b. Alternately OLCF would provide a GitHub account that would be accessible to all OLCF users.
c. Alternately, Use ITSD’s code.ornl.gov or gitlab.ornl.gov – these areas are only open to people with ORNL XCAMS or UCAMS – so not all user would have access.
d. Users will and have set up open code repository specialized for Titan. The question is how or if OLCF can facilitate or advertise such repositories so that users can collaborate and more broadly benefit from them.
2. Give each project a webpage where they can briefly describe their project, show-off results, post information about code resources that they have available, and post areas of expertise where they would be interested in collaborating.
3. Give projects time during the user call to outline how they have tackled HPC programming tasks. Users could share links to code. The slides could be posted on the events page with or without links.

Discussion

The primary issues for OLCF with hosting or linking to open access user generated content are related to ensuring that export controlled material is not posted or advertised by OLCF. There is a gain for OLCF in facilitating this kind of User Collaboration if it can be done without a risk of violating export control. The ideas above will be reviewed for risk and or feasibility of a review process.

OLCF currently has a GitHub account for posting training material but this account is limited to only a few projects and only staff members can post to it, however users have contributed staff vetted content. There also was discussion of the burden of maintaining User Driven code repositories. GitHub does not have inherent organization strategies for maintaining a large number of codes in one repository.

User Survey Participation Request:

Balint will write a draft of a message from the OLCF User Group Ex. Board (OUG EB) asking users to participate in the 2014 OLCF User survey. The Board will review this message and send it to OLCF by Oct. 31st for distribution.

General User Group Call

30 Users and Staff attended this call, 20 by WebEx, 10 in person.

Luster2.5
The recent update to Lustre 2.5 clients on Titan, Rhea, and Eos was discussed.
A few user tickets discussing recent performance observations were mentioned by users. OLCF is following up. (Action item taken by user assistance and outreach (UAO))

CLE5.2 UP02
The upcoming upgraded to CLE 5.2 up02 was also discussed. The upgrade will take place in late 2014 or early 2015. This update will require users to recompile and relink their applications because newer versions of key tools and lib. are required by the update (MPT).

Users requested access to the aforementioned tools, specifically cce.8.3.0. Cce.8.3.0 is currently on Titan but it requires cray-mpich2/7.0.0, which is not currently on Titan. Cray and UAO are following up to see what can be made available. (Action Item taken by Cray and UAO)

Two users requested that we make CLE5.2up02 available as soon as possible because of needed performance increases it delivers with MPI and the need for features in CLE5.2up20 dependent Cuda6.5 that are not supported with our current Cuda. Cray Cuda6.5 will be available in December 2014 at the earliest. The follow-up ticket sent by the user to the OLCF help line is #219320.

A request was made for information regarding when the new version of Cuda Fortran will be available. The release of Cuda Fortran is dependent on the releases of the PGI compiler. The updated version will not be available before Cray CUDA6.5 is released and installed.

A user also reported missing data in $MEMBERWORK. We requested the user to send a ticket to the UAO help line. This was done and we are following up to determine if this is a result of the regularly scheduled sweep for files older than 14 day sitting in $MEMBERWORK. The filesystem performance is dependent on the space liberated by the sweep. Note that the sweep interval for $PROJWORK is 90 days and users have access to the hpss archive which is not swept.