DPM dev enlarged meeting

Friday 30 Aug 2013, 10:00 → 12:00 Europe/Zurich

28/R-015 (CERN)

 

 

 

- push from EPEL test to EPEL will likely happen next week, as Cristina will be back

  - we would like to monitor a prod site that will be the first one to upgrade. Who's available?

i will get in touch with cristina and sync for wed/thu next week.

  * Wahid will update first.

 

 

- progress on Ganesha/NFS

  - do we need to do anything? ShuTing said that they would have liked to use our build system

  * ShuTing: they can start, and don't know how to install, i will give info to shuting if

    needed. First thing is that ShuTing should to see if they have access to the Ganesha and dpm-nfs SVN at CERN.

    Otherwise will need to give her access to our svn. We can get in touch by email to do

    that.

 

 

- we started running little, regular HC perf tests towards sites. The perf of HTTP towards CERN improved, will improve more

  * progress on Fabrizio's TTreeCache tests/debug. After investigation, Wahid discovered

    that the script for HTTP and for Xrootd were different and behaving differently

    Wahid should change the tests to using the same script, if possible. The situation is difficult because of the

    workarounds that are needed to make TWebFile kind of work. 

    (news: he has fixed them, now they are the same) fab: which platform runs the clients? varies sl5/sl6.

     Fab: likely we will start with testing davix next week, as I just set up my new dev environment for ROOT things

 

     * Wahid will fix the config of his SE, enable keepalive. Martin: you should also check for deep inspecting

       firewalls that may impact performance. Wahid: will see.

 

- announcement: end of support of YAIM is Aug 2014. Puppet is the way.

  * no comments.

 

- admin tools

  * Sam is restarting to have a look at them.. will have a look at the dmlite calls. having a look at andrey's db checking tool.

 

- progress on VFS

  * nothing blocking. all issues clarified, not yet implementation available.

 

- AoB

Sam: interested in the rebalancing. Other sysadmins saw load spikes similar to Glasgow. What about load limitation ?

Fab: load limitation has a boundary with denial of service. Sam: the issues happen with rfio, an idea would be

moving to xrd or http. Fab: maybe somebody did forms of throttling for http as well?

Sam has found that they can move the problem passing to xrootd.

Will be nice to implement something like a queueing manager for http and dpm. Will need a

client that supports the 'wait and retry' response of http.

Martin: could be nice to do the rebalancing.

Fab: difficult to determine when to start a proactive rebalancing, as it would add load.

Martin: we should find a way to do the rebal without adding load in the wrong moment

Sam has a pres for chep that talks about rebalancing tools.

Main issue is that there is no target replication in dpm, as he needs to write to specific

pools or to specific filesystems in pools. Fab: see no issues in planning to add such a thing.