Re: cgroupfs, /hurd/proc and subhurds
Quoting Ludovic =?utf-8?Q?Court=C3=A8s?= (2013-09-17 13:47:23)
> Hi!
>
> Justus Winter <4winter@informatik.uni-hamburg.de> skribis:
>
> > Linux has this feature called cgroups. It groups processes (threads)
> > together in groups, furthermore so called controllers can be used to
> > restrict the use of various resources (like cpu time, memory) on a
> > per-group basis.
>
> Isn’t restricting resource use very difficult in practice due to the
> fact that servers allocate resources on behalf of clients? (As noted in
> the “Critique”.)
>
> For instance a process within a “cgroup” could use the io interface of a
> server outside the “cgroup” and cause that server to spend lots of
> memory and CPU time serving those requests.
It is difficult indeed. I just mentioned one facet of how cgroups are
used on Linux. cgroups are useful without it, as they make tracking
child processes possible. systemd uses this to monitor services and to
reliably kill services.
Restricting the cpu time based on cgroups could still be useful, even
if not as effective against DOSing servers. It is after all
(hopefully) not possible to make a server do arbitrary computations on
behalf of a client.
Also, finishing the thread migration in gnumach could make proper
accounting of (some) resources possible.
Justus
Reply to: