Minutes of the storage extralong EVO meeting 13 Oct 2010

Present:
	Bristol: Winnie
	Sheffield: Elena
	Liverpool: John, Stephen
	Edinburgh: Wahid
	QMUL: Chris
	Lancaster: Matt
	Glasgow: Sam, David
	RAL: Jens (chair+mins)


1.

Jens talked about the demonstrations at the pre-GDB (see link in chat
log).  Brian had prepared some data for an ATLAS file (because work has
to be demonstrated with real experiment data), demonstrating that getting
the file (or rather the TURL) took about 5-6 seconds for most SEs, but
only about 1 second for BeStMan.  It still is not instantaneous because
there is a check that the file is really present on the disk server.

2a.

Wahid gave presented the latest version of his CHEP talk.  See chat
window for link to the presentation (presumably it will eventually
be available at the CHEP agenda page as well)

http://indico2.twgrid.org/conferenceTimeTable.py?confId=3

2b.

Sam gave a presentation of the SSD work - see chat window for a link to
the presentation.

In general, if people have comments about the presentations, could they
please contact Wahid/Sam (as applicable) directly, or discuss on the list.

CHAT

[09:59:15] Winnie Lacesso Can we download these presentations i.e. pdf or something?
[09:59:56] Wahid Bhimji joined
[10:00:27] Wahid Bhimji http://www2.ph.ed.ac.uk/~wbhimji/GridStorage/Wahid-Chep.pdf
[10:00:48] Stephen Jones joined
[10:01:06] Jens Jensen http://indico.cern.ch/conferenceDisplay.py?confId=72063
[10:02:16] Christopher Walker joined
[10:03:16] Matthew Doidge joined
[10:03:22] Sam Skipsey joined
[10:04:39] David Crooks joined
[10:08:59] Jens Jensen BTW I believe Brian is currently listening to the GDB
[10:09:46] Sam Skipsey http://svr001.gla.scotgrid.ac.uk/CHEP2010-SSDtalk.pdf
[10:26:25] Christopher Walker p5 Check spelling of environment
[10:29:19] Jens Jensen We've lost you...
[10:31:07] Elena Korolkova which rfio buffer size is better?
[10:31:25] Elena Korolkova I 've missed the number, sorry
[10:33:56] John Bland sorry, just had vendor on phone
[10:37:23] Wahid Bhimji john/chris/others -if you have any comments do send them to me. 
[10:37:38] Wahid Bhimji Sorry I'm a bit sleepy - baby not sleeping well at the moment
[10:38:24] Ewan Mac Mahon joined
[10:39:28] John Bland can whoever's typing stop it or mute their mic, please?
[10:40:52] Elena Korolkova As I think my question is lost i will repeat it: what is best RFIO buffer which is recommended to use ATM?
[10:42:18] Sam Skipsey   User's echo suppression automatically activated
[10:44:55] Wahid Bhimji Elena - sorry there is no definate conlcusion sadly. Smaller buffers 4k are definately better for direct rfio access IF it doesn't overload your servers. But as most ATLAS jobs copy to WN and servers do get overloaded by random access jobs - so a compromise is needed
[10:44:59] Wahid Bhimji probably 0.5 MB
[10:46:51] Elena Korolkova Many thanks for explicit answer, Wahid
[10:55:52] John Bland gotta go, meeting about money!
[10:56:00] John Bland left
[10:57:09] Christopher Walker Of course the metric should be performance per total server price (which won't change your conclusions) 
[10:57:15] Christopher Walker Got to go group meeting
[10:57:52] Christopher Walker Too late to suggest tmpfs would be interesting to test
[10:58:39] Matthew Doidge Another nail in the coffin for SSDs would be the atlsa requirement for 50GB of free space/job
[10:58:44] Matthew Doidge *atlas
[11:04:20] Elena Korolkova Wahid, could you increase atlas hot disk
[11:04:29] David Crooks left
[11:04:30] Ewan Mac Mahon left
[11:04:32] Matthew Doidge left
[11:04:34] Winnie Lacesso left
[11:04:42] Stephen Jones left
[11:05:07] Elena Korolkova left