 |
 |
On 04/05/12 12:36, Alessandra Forti wrote: > Isn't that equivalent to giving a 50-50 (assuming 2 CEs) weight to the > different HS06 independently from the number of nodes? It doesn't seem > really correct thing to do. Mathematically you are correct. However, it means that the grid does the maths[1]. I can say I have A nodes with B hepspec and C nodes with D hepspec. When I add some additional nodes, I can easily change it without digging out the spreadsheet. In some sense, I think it's also interesting to publish the nodes we have, rather than some weighted average. To answer Stephen's point, jobs can't rely on which node they end up on - so job requirements might not be met. However, the nodes we have have fairly similar hepspec scores, and RAM/jobslot. Clearly they will have different RAM/node, and indeed different MPI capabilities - but the former is pretty irrelevant for the job, and the latter isn't currently used. I'm pretty sure that's the way most people run their grid sites - apart from anything else, it avoids jobs going to machines you've switched off. In summary, I think it's a point of style, and won't make any difference in practice. I'm willing to be puersuaded otherwise. Chris [1] I'm sure that judicious use of expr would enable the yaim config file to do the maths, but I haven't got around to coding that yet. > > On 04/05/2012 12:16, Christopher J.Walker wrote: >> On 04/05/12 12:05, Jeremy Coles wrote: >>> Dear All >>> >>> You can check figures via >>> http://gstat2.grid.sinica.edu.tw/gstat/summary/Country/UK/ (though >>> UCL seems to be missing). >>> >> QMUL has 3840 jobs running at the moment. >> http://gstat2.grid.sinica.edu.tw/gstat/summary/Country/UK/ reports 10437 >> >> However looking at https://ggus.eu/ws/ticket_info.php?ticket=64388 >> that's probably because I use different CEs to publish the different >> node types. This means I don't have to worry about a site average score >> when publishing hepspec scores. >> >> It would be possible to change this if desired. >> >> Chris >> >> >> >>> Jeremy >>> >>> On 4 May 2012, at 12:01, Stephen Burke wrote: >>> >>>> Hi, >>>> >>>> There’s a long-standing problem that gstat overcounts things like >>>> RunningJobs and TotalJobs if there are multiple CE nodes feeding the >>>> same set of queues. The gstat developers have just put a comment on >>>> my ticket for that indicating that they have done something to >>>> remove the duplicates: >>>> >>>> https://ggus.eu/ws/ticket_info.php?ticket=64388 >>>> >>>> Could sites who are in that situation have a look to see if the >>>> numbers are OK (or at least better) now? >>>> >>>> Stephen >>>> >