Throughput Meeting --- October 27, 2016
Attended: Shawn, David, Enrico, Saul, Frederique, Frederic, Andy, Duncan, Garhan, Marian
0) Agenda review/update, News
Shawn reported on WLCG workshop network session, HEPiX and CHEP.
WLCG networking session at San Francisco workshop:
Quite a few questions on what we need from the experiments, how we move forward, further discussions planned at the pre-GDB on networking in January 10th at CERN (https://indico.cern.ch/event/571501/)
CHEP 2016 networking plenary talks:
HEPiX Security and Networking sessions:
1) perfSONAR/infrastructure status
Duncan reported an issue with dashboards not updated after mesh config change, Shawn fixed the maddash.aglt2, but we need to also fix the OSG production ones.
Andy reported perfSONAR 4.0 RC2 should be released on Monday (31st Oct), bunch of issues got fixed, mainly with pScheduler. Testbed instances not showing up any results, to be reviewed again once RC2 is deployed.
Marian proposed to setup abrtd reporting on the testbed nodes to get auto-generated reports on seg. faults. Andy said the best way to report bugs is directly to the ps-dev mailing list, Marian to create a list to receive automated abrt mails and forward what looks important.
Marian noted CERN ITB message broker changed to new configuration on VM/CC7 (previously was physical machine on SL6) - same ActiveMQ version is used.
Shawn working on packaged version of the new mesh-configuration and getting additional funding for Soichi to support bug-fixing and move this to production. The plan is to do this along the pS 4.0 campaign, we plan to have transparent transition to the new configuration.
Duncan proposed to extend IPv6 meshes to OPN, USCMS and UK mesh. Marian mentioned that we could move to default testing (which favors IPv6 over IPv4 if both are available), but we would loose the possibility to compare the results.
Saul's optimizer to be implemented in the FTS 3.5.x release :-) (https://indico.cern.ch/event/564202/), nb. FTS will be focused topic at the next WLCG operations coordination meeting (https://indico.cern.ch/event/540423/)
3) Focus topics: pre-GDB and Network Analytics
Marian presented planned areas for the pre-GDB network session, we're soliciting input on what areas/topics should be presented. This was followed by a summary of the network analytics done at CERN (slides attached to indico -> https://indico.cern.ch/event/577645/)
Ilija commented that neural network training could be improved and there was some discussion on the jupyter code and how we could re-deploy a new NN to production. There could be some effort on this at Univ. of Chicago soon.
4) Round-table: Items to discuss?
Frederic reported on tracking down IPv6 outbound performance issues in the French cloud (ticket still open is at https://ggus.eu/?mode=ticket_info&ticket_id=123930; looks like asymmetric path)
Saul reported that they should be joining LHCONE soon.
Garhan reported packet loss on Nebraska to Fermilab path, but couldn't replicate from command line, to be followed up.
David reported an issue with one of MWT2 router interfaces, causing significant loss btw Chicago and Indiana, faulty interface was replaced. Also, asked if there is way to get notified of such issues, Ilija will send a link for the notification/alerting subscription mechanism to the mailing list (feedback at the next meeting).
5) AOB and next meeting: NA meeting after SC16