mbuf leak with SMP and debug.mpsafenet=1
Andrew Gallatin
gallatin at cs.duke.edu
Tue Oct 19 14:29:48 PDT 2004
Robert Watson writes:
> On Thu, 14 Oct 2004, Andrew Gallatin wrote:
>
> > Log into another host and do:
> >
> > % while 1
> > while? ssh 'netstat -m | grep mbufs'
> > while? netperf224 -Hscream -tUDP_STREAM -l60 -- -m 1 >& /dev/null
> > while? end
>
> I must have missed this e-mail previously, sorry about the slow response.
> John Baldwin pointed me at it wondering why I hadn't answered, and that's
> why :-).
Funny you should write now. I noticed green's MFC of a fix for mbuf
leaks and was trying RELENG_5 from this morning. Alas, it does not
fix my problem.
> Is the number of mbufs lost proportional to total run time since the
> start, or number of runs since start? I.e., if you shorten 60 seconds to
> 6 seconds but sample at the same rate, what impact does that have on the
> measured leakage?
I just tried that and it looks like the leakage stays rougly the same.
It seems to be roughly 100-150 mbufs/second with this setup.
I hooked up the em0 GbE interfaces, and that leaks nearly as bad as
my myrinet nic (at least with a linux sender, hooked back-to-back).
Em0 seems to be leaking at a few thoundsand pkts/sec, so I wasn't brave
enough to do a long run..
Drew
More information about the freebsd-current
mailing list