 |
 |
Folks, Having spent nearly two years working with SA3 and JRA1 on stress- testing the WMS, I'd be the first to say that it isn't perfect. Even now, with all of the experience I've gained, I find that the WMS can sometimes get itself into a state where only the most brutal of interventions is enough to save the day. Unfortunately, this can also lead to jobs being lost, and this is obviously unacceptable. Having said that, we are in a much better position than we were even twelve months ago, especially in respect of bulk submission. Returning to my original question, the one point that stood out for me was Martin's figures for CPU power and memory. The existing WMS can be astonishingly demanding on processor and memory (insofar as I can see, it will take as much as you give it), and this is at least part of the key to keeping the middleware happy. I'm aiming (resources permitting) to go with something like the following: WMS 500GB drive 16GB RAM Dual quad-core Intel Xeon at 3GHz LB 500GB drive 8GB RAM Single quad-core Intel Xeon at 3GHz I'm also installing a CREAM CE, where the pilot studies suggest that the following should be adequate: 500GB drive 8GB RAM Dual quad-core Intel Xeon at 3GHz All the best, Barry. On Fri, 13 Feb 2009, Gordon, JC (John) wrote: > Date: Fri, 13 Feb 2009 15:08:53 -0000 > From: "Gordon, JC (John)" <[log in to unmask]> > Reply-To: Testbed Support for GridPP member institutes > <[log in to unmask]> > To: [log in to unmask] > Subject: Re: Specs for latest WMS & LB > > Yes, everything in the world is interconnected but that doesn't mean > every discussion is completely open. Barry specifically asked about > hardware specs and it was you that raised the point about it not being > the limiting factor. By all means let's have a discussion about the > problems with WMS but can we have a new thread please? > >> -----Original Message----- >> From: Testbed Support for GridPP member institutes [mailto:TB- >> [log in to unmask]] On Behalf Of Graeme Stewart >> Sent: 13 February 2009 14:24 >> To: [log in to unmask] >> Subject: Re: Specs for latest WMS & LB >> >> I presume he said hardware because he wants to offer a service to >> users. The fact that users may not have a good experience even if that >> hardware is excellent is surely relevant to this discussion. >> >> Graeme >> >> On Fri, Feb 13, 2009 at 3:15 PM, Gordon, JC (John) >> <[log in to unmask]> wrote: >>> Come on Graeme, he said 'hardware'. You yourself reported earlier >> today >>> from the GDB that WMS problems did not appear to be > hardware-related. >>> >>> John >>> >>>> -----Original Message----- >>>> From: Testbed Support for GridPP member institutes [mailto:TB- >>>> [log in to unmask]] On Behalf Of Graeme Stewart >>>> Sent: 13 February 2009 14:01 >>>> To: [log in to unmask] >>>> Subject: Re: Specs for latest WMS & LB >>>> >>>> Maybe Catalin is happy, but the users are not... >>>> >>>> We spent sometime debugging stalled jobs from a Lancaster user this >>>> week, which had been fed through the RAL WMSs. >>>> >>>> Steve's tests have another nice red stripe down them today: >>>> >>>> http://pprc.qmul.ac.uk/~lloyd/gridpp/atest.html >>>> >>>> "Test Job Result for MyAnalPackage at 13 Feb 2009 09:32:47 >>>> Status Failed >>>> Reason Job submission timed out >>>> CE Node gridgate.cs.tcd.ie >>>> Total time -1 >>>> Result >>>> Submit Time -1 >>>> Status Time -1 >>>> Output Time -1 >>>> >>>> Submission Output: >>>> >>>> glite-wms-job-submit -a --config lcgwms01.gridpp.rl.ac.uk.conf >>>> temp_UKI-IRELAND-TRINITY_MyAnalPackage.jdl" >>>> >>>> Cheers >>>> >>>> Graeme >>>> >>>> On Fri, Feb 13, 2009 at 2:55 PM, Bly, MJ (Martin) >>>> <[log in to unmask]> wrote: >>>>> Catalin reports he is happy with our WMS/LB hardware... >>>>> >>>>> Martin. >>>>> -- >>>>> Martin Bly >>>>> RAL Tier1 Fabric Team >>>>> >>>>>> -----Original Message----- >>>>>> From: Testbed Support for GridPP member institutes >>>>>> [mailto:[log in to unmask]] On Behalf Of Bly, MJ (Martin) >>>>>> Sent: 13 February 2009 13:25 >>>>>> To: [log in to unmask] >>>>>> Subject: Re: Specs for latest WMS & LB >>>>>> >>>>>> FYI, our three WMS and two LBs are running on the following: >>>>>> >>>>>> Dual chip quad-core xeon E5410 (2.33GHz) >>>>>> 16GB RAM >>>>>> 2 x 500GB enterprise SATA in software RAID1 >>>>>> >>>>>> I've asked Catalin whether he is happy with it but no >>>>>> response yet (it's >>>>>> lunch time) - but he hasn't complained since we started using >> that >>>>>> config! >>>>>> >>>>>> Regards, >>>>>> Martin. >>>>>> -- >>>>>> Martin Bly >>>>>> RAL Tier1 Fabric Team >>>>>> >>>>>>> -----Original Message----- >>>>>>> From: Testbed Support for GridPP member institutes >>>>>>> [mailto:[log in to unmask]] On Behalf Of Dr Barry >> MacEvoy >>>>>>> Sent: 13 February 2009 11:04 >>>>>>> To: [log in to unmask] >>>>>>> Subject: Specs for latest WMS & LB >>>>>>> >>>>>>> Folks, >>>>>>> >>>>>>> What are the recommended machine specs for the latest WMS & LB >>>>>>> (i.e. the versions in certification) if you want to get the >> VERY >>>>>>> BEST out of the middleware under heavy load? >>>>>>> >>>>>>> From the tests I've performed over the last 12 months, I'm >>>> guessing: >>>>>>> >>>>>>> 8GB RAM (maybe just 4GB for the LB) >>>>>>> 500Gb disk or greater >>>>>>> Single Quad-core Intel Xeon ~3GHz or better >>>>>>> >>>>>>> Does that sound reasonable? >>>>>>> >>>>>>> Please let me know, as I'm intending to deploy an additional >>>>>>> WMS and LB in London for very heavy use and to retire my >> existing >>>>>>> WMS and LB to the role of backup. >>>>>>> >>>>>>> Cheers, >>>>>>> >>>>>>> Barry. >>>>>>> >>>>>>> >>>>>>> -------------------------------------------------------------- >>>>>>> Dr Barry MacEvoy >>>>>>> High Energy Physics Group >>>>>>> Imperial College London >>>>>>> Blackett Laboratory >>>>>>> Prince Consort Road >>>>>>> LONDON SW7 2BW >>>>>>> England >>>>>>> >>>>>>> T: +44 20 7594 7802 >>>>>>> F: +44 20 7823 8830 >>>>>>> M: 07767 323871 >>>>>>> >>>>>>> http://www.hep.ph.ic.ac.uk/cms/people/based_at_imperial.html >>>>>>> http://www.hep.ph.ic.ac.uk/e-science/people/macevoy.html >>>>>>> -------------------------------------------------------------- >>>>>>> >>>>>> -- >>>>>> Scanned by iCritical. >>>>>> >>>>> -- >>>>> Scanned by iCritical. >>>>> >>>> >>>> >>>> >>>> -- >>>> Dr Graeme Stewart > http://www.physics.gla.ac.uk/~graeme/ >>>> Department of Physics and Astronomy, University of Glasgow, > Scotland >>>> DEATH TO MEETINGS! >>> -- >>> Scanned by iCritical. >>> >> >> >> >> -- >> Dr Graeme Stewart http://www.physics.gla.ac.uk/~graeme/ >> Department of Physics and Astronomy, University of Glasgow, Scotland >> DEATH TO MEETINGS! > -- > Scanned by iCritical. > -------------------------------------------------------------- Dr Barry MacEvoy High Energy Physics Group Imperial College London Blackett Laboratory Prince Consort Road LONDON SW7 2BW England T: +44 20 7594 7802 F: +44 20 7823 8830 M: 07767 323871 http://www.hep.ph.ic.ac.uk/cms/people/based_at_imperial.html http://www.hep.ph.ic.ac.uk/e-science/people/macevoy.html --------------------------------------------------------------