On Wed, 25 Dec 2013, Potapov Sergey wrote: > Hi! > > I`am accessing to Ceph cluster via librados and found that memory usage > (VSZ in ps) is extremely increases on writing/reading/removing objects. > If I successively write+read+remove the object with the same name, then > memory usage is not increased, but if I do > the same operations with the different object names then VSZ increases > up to 4G and stops at this level. It seems that some internal buffering > is enabled, can I configure the maximum size of them? Memory usage is > very critical for me and I want to reduce it as much as possible. > > Operations write+read+stat+remove in loop with different object names: > lion@kubuntu-12:~$ ps aux | grep RADOS > lion 7890 0.2 0.0 4289264 3988 ? Sl 10:28 0:00 /home/lion/Projects/RMS/TestRADOS/TestRADOS > > Operations write+read+stat+remove in loop with one object name: > lion@kubuntu-12:~$ ps aux | grep RADOS > lion 8149 0.1 0.0 556424 3520 ? Sl 10:30 0:00 /home/lion/Projects/RMS/TestRADOS/TestRADOS > > Listing of test program: http://pastebin.com/idqt1PEv There will be some memory usage that is proportinoal to the size of the cluster because of the sockets open to the various OSDs. I'm not sure what level you should expect, though: it depends on the size of the cluster, your architecture...and I haven't really measured it. In any case, if you can run your program through valgrind massif, that will tell us exactly where the memory is going, and whether there is anything obviously wrong or easy to fix! sage _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |