[Halld-online] Online Meeting Minutes

David Lawrence davidl at jlab.org
Thu Jun 16 09:40:43 EDT 2016


Hi All,

  Minutes from yesterday’s Online Meeting are now posted at the wiki page and copied below.

https://halldweb.jlab.org/wiki/index.php/OWG_Meeting_15-Jun-2016

Regards,
-David

Minutes[edit]

Attendees: David L., Sergey F., Simon T., Dave A., Carl T., Vardan G., Sean D., Beni Z.
Announcements[edit]

Whiteboard cameras are now both working and posting pictures every 10 minutes to web. Cue username and password are required to access the page.
Computer/Networking[edit]

Most items from above agenda are self-explanatory
gluon40 IB card was preventing system from booting. Paul L. exchanged with one from gluon43 before sending it out so that gluon40 is up and working again.
Computers in counting house are not aligned in OS versions. David is going to check with Kelvin on what it will take a to align them all to RHEL6.5 Schedule will be announced once it is formed.
There was some discussion on whether we should consider a fast ethernet switch for the new switch purchase rather than infiniband
Sergey pointed to a few motivating factors:
Ethernet drivers automatically built into Linux kernel and don't need to be managed externally like infiniband
Tests on Hall-B 40GB ethernet showed 27GB/s for single iperf instance whereas our current infiniband configuration requires 2 iperf instances (2 links) to obtain that speed
Price difference between fast ethernet and infiniband has been dropping so may not be as big of an issue as it once was. Particularly in the future since ethernet's popularity will likely continue to drive down prices
Dave A. pointed to a few motivating factors:
Our current Infiniband configuration uses IP over IB and is not optimal for reaching full bandwidth of hardware
Mellanox hardware can be configured to use fast ethernet (driver will still be outside of kernel, but could then talk to another ethernet device)
Infiniband is used extensively in HPC environments (including JLab's) and has libraries/features that can maximize throughput by bypassing overhead of networking protocols like IP.
Price differential between 40GB ethernet and 40GB IB is still significant (Sergey thought when he'd looked it wasn't so big)
Decision was made to go forward with a new IB switch. (n.b.Sergey maintained a dissenting opinion, but only slightly)
Default compiler and python versions[edit]

hdonline.cshrc was modified to set different defaults for gcc compiler and python versions based on which computer is being logged into
gluon40 and higher plus the gluon RAID servers now setup environment to use gcc4.9.2 and python 2.7.11 as default
gluon01-gluon39 remain unchanged
Environment can be easily changed back and forth using he "setup" utility. (Just type "setup" with no arguments to see list of options)
DAQ[edit]

Most crates in hall remain off so full testing of CODA 3.0.6 is being deferred until they are turned back on.
Crates are getting turned back on gradually by Alex S. as he installs and tests new firmware versions
Front-end Firmware Status[edit]

No report (Alex S. not at meeting)
L3 Status[edit]

Effort is still ramping up. Benchmarking various parts of code is underway
David L. showed some slides from last week's L3 meeting indicating some odd behavior where the event processing rate initially rises to 12kHz and then drops to 8kHz.
Suggestion was to try looking for optimal configuration regarding number of parsing threads and processing threads. (This mirrored a suggestion from the L3 meeting. David L. will do this.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-online/attachments/20160616/b909f27d/attachment.html>


More information about the Halld-online mailing list