The meeting date/time is a result of a fixed schedule agreed in the last meeting (https://indico.cern.ch/event/354593/).
Details on the Network and Transfer Metrics WG are available at our Twiki.
Network and Transfer Metrics WG (6th May April 2015) Minutes
Attended: Hung-Te Lee, Henryk, Alessandro DiGi, Bruno, Stefan R, Shawn, Tony, John, Saul, Tomas, Michael, Jason, Julia, Marian
Agenda/slides presented at https://indico.cern.ch/event/382623/
Next meetings: 3 June, 8 July, 2 Sept, 7 Oct - all at 4pm CEST
1) CHEP paper
Deadline is 17th of May. It was proposed we work on it the same way as for the use cases document, via Google Docs. The link to the document is:
Please let us know references to cite and acknowledgements to mention. It was also proposed we don’t add affiliations for all authors, since this would take the entire page. We’ll send a draft version for review next week, please add comments and/or proposed changes directly in Google Docs. Final formatting for submission will be done in Latex (so you can use latex markup if needed).
2) perfSONAR status
Good progress in the latency ramp up, now with 34 sonars, seeing 95 % of results on a consistent basis. Few changes in ALL meshes, we moved from iperf to iperf3 and from traceroute to tracepath. News on security and infrastructure monitoring were presented.
Please register in OIM to use our configuration interface at https://oim.grid.iu.edu/oim/meshconfig , please contact me or Shawn if you have issues.
Current deployment status (http://grid-monitoring.cern.ch/perfsonar_report.txt):
WLCG perfSONAR service status report on 2015-05-07 04:02:24.706444 =======
Active perfSONAR instances: 235
Registered/monitored perfSONAR instances: 259
perfSONAR-PS versions deployed:
3.4.1 : 23
3.4.2 : 183
Incorrectly configured (failing >4 metrics): 17
Mesh leaders, please check status of sonars still on 3.4.1 in your region (they either run out of disk or have auto-updates disabled, which poses a potential security problem and is not recommended). Please note that we strongly recommend to keep auto-updates enabled.
3) Network Incidents Follow up (nothing to report)
4) Datastore/esmond status (detailed report planned on 8th of July)
5) Pilot projects -
Experiments Interface to perfSONAR (Henryk has sent an update via e-mail)
Hassen reported on FTS performance for WLCG during the first phase of production (3 months). Saul asked what determines the number of tcp streams and Michael answered that there are three different modes for this (based on file size, probe and experimental) and explained the differences.
Initial report on the FTS performance study was presented by Saul, common for ATLAS, CMS and LHCb. This was followed by discussion on the tcp buffers and tcp streams. Costin commented that in Alice they recommend to set tcp buffers to 8MB for all nodes. Shawn commented that newer version of Linux usually have auto-tuning enabled for this, but this is not the case for SLC (which also defaults to a very low number), Costin also noted that it’s quite complicated to get all sites to change this (since it needs to be done on all nodes). It was agreed to follow up on the Saul’s work on a regular basis at the next meetings.
Next meeting will be focused on latency ramp up (mesh changes) and proximity service (June 3rd 4 PM CEST, https://indico.cern.ch/event/382624/).