I don't currently see any. It the case of the account gather you not only have the overhead of the cgroup reading but you also have the reading of proc since the cgroup doesn't contain all the stats you need, only a very small subset. So it just takes longer. The proctrack plugin is very fast I highly recommend using it, just not the account gather. The task plugin also adds overhead as well depending on the cgroups you use but most people won't notice. You would have to be running 10s of jobs a sec to really notice. It dies add quite a bit of functionality so it isn't something I generally discourage.
On December 23, 2015 3:40:28 PM PST, Christopher Samuel <[email protected]> wrote: > >Hiya Danny, > >On 23/12/15 14:41, Danny Auble wrote: > >> We do use cgroups to get the processes, at least if you use the >cgroup >> proctrack plugin. The only cgroup plugin I usually don't suggest is >the >> account gather one. I don't believe scalability is an issue when done >> this way. > >What scalability issues do you see with the cgroup plugin, and what >causes them? > >Does it take longer to read from cgroup filesystems? > >All the best! >Chris >-- > Christopher Samuel Senior Systems Administrator > VLSCI - Victorian Life Sciences Computation Initiative > Email: [email protected] Phone: +61 (0)3 903 55545 > http://www.vlsci.org.au/ http://twitter.com/vlsci
