>From my experiences and understanding, GridEngine tracks how much memory is in use and held by the entire task. So, if part of the task has run, but has not deallocated the RAM it used, GE still considers that unused but still allocated RAM as part of the entire job overhead.
But, I could be wrong. Ian On Tue, Apr 15, 2014 at 4:46 AM, Lars van der bijl <[email protected]> wrote: > hey Reuti, > > so I followed all PID's created over the life of the task. > > Name: _clue_runner Pid: 32643 Memory: 309542912 Memory human: 295.2 MB > Name: mv Pid: 32709 Memory: 116514816 Memory human: 111.1 MB > Name: mv Pid: 390 Memory: 116514816 Memory human: 111.1 MB > Name: sh Pid: 32650 Memory: 108810240 Memory human: 103.8 MB > Name: hython-bin Pid: 32652 Memory: 854700032 Memory human: 815.1 MB > Name: mv Pid: 301 Memory: 116514816 Memory human: 111.1 MB > Name: mv Pid: 367 Memory: 116514816 Memory human: 111.1 MB > Name: mv Pid: 32721 Memory: 116514816 Memory human: 111.1 MB > Name: mv Pid: 32756 Memory: 116514816 Memory human: 111.1 MB > Name: mv Pid: 314 Memory: 116514816 Memory human: 111.1 MB > Name: mv Pid: 349 Memory: 116514816 Memory human: 111.1 MB > Memory Total: 2205171712 2.1 GB > > this seems to be adding up to about what i've been seeing. now my question > is, does gridengine remember all PID's over the lifetime of the submission > and aggregates that? > > > > On 14 April 2014 16:40, Reuti <[email protected]> wrote: >> >> Hi, >> >> Am 14.04.2014 um 16:20 schrieb Lars van der bijl: >> >> > over the last few weeks we have been having some problems with s_vmem >> > and array tasks. >> > >> > we submit a task with a memory requirement s_vmem of 2G. >> > it starts running and I follow it on the machine using pmap >> > >> > [root@atom08 ~]# pmap -x 19893 >> > hythonRender -f 80 170 -i 1.0 /out/MANTRA1/rs_mantra_pipe/ifd_gen >> > rnd_lars_main_v001_lars_none.hip_out_ALL_2014-04-14_11.58.21.hip >> > Address Kbytes RSS Dirty Mode Mapping >> > 0000000000400000 28 24 0 r-x-- hython-bin >> > ...... >> > ffffffffff600000 4 0 0 r-x-- [ anon ] >> > ---------------- ------ ------ ------ >> > total kB 844844 191828 117452 >> > >> > so that tells me the task is taking about 840M of memory. >> > >> > If I then have a look at the memory usage in the qacct >> > >> > $ qacct -j 120429 >> > .. >> > cpu 17.302 >> > mem 10.299 >> > io 2.726 >> > iow 0.000 >> > maxvmem 1.979G >> > >> > which tells me it was 1.9G. sometimes its more like 2.1G and then the >> > s_vmem gives the signal. >> > is there something I'm missing as to how maxvmem is calculated? >> >> SGE will not look at one process only, but at the complete process tree. >> Is PhythonRender the one and only child of `sge_shepherd`? >> >> -- Reuti > > > > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users > -- Ian Kaufman Research Systems Administrator UC San Diego, Jacobs School of Engineering ikaufman AT ucsd DOT edu _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
