 |
 |
Ewan MacMahon wrote: >> -----Original Message----- >> From: Testbed Support for GridPP member institutes [mailto:TB- >> >> Suggestion is 5MB/s per analysis job. What the error on this value is; >> well........ >> > Hmm. That clearly assumes a constant performance per CPU core, > but it's probably good enough for what I'm going at the moment. > > It would be nice to firm that up though. > I think the numbers I gave in the talk at RHUL were: An event is 200 kB in size, and Atlas used to expect 10Hz, giving a data rate of 2MB/s Now that the AOD files are a lot better, one might assume 25Hz and 5MB/s This assumes files are only transferred once (ATLAS are transferring them twice at QMUL[1]). For 12 jobs on a 24 thread machine, thats around 60MB/s or probably a bit over half a gigabit. I can't remember where I got the 200kB per AOD event. I'm sure the actual number has changed. Chris [1] First they checksum all the files for analysis, then the read the first one. This of course ensures that the file is long gone from the page cache by the time it is read.