Date: Fri, 8 May 1998 09:59:19 +0000 (GMT) From: Chris Fanning <cfanning@jingoro.prevmed.sunysb.edu> To: P.Gevros@cs.ucl.ac.uk (Panos GEVROS) Cc: dag-erli@ifi.uio.no, freebsd-net@FreeBSD.ORG Subject: Re: MBUFs and IPFW revisited Message-ID: <199805080959.JAA12235@jingoro.prevmed.sunysb.edu> In-Reply-To: <29428.894629525@cs.ucl.ac.uk> from Panos GEVROS at "May 8, 98 01:12:05 pm"
next in thread | previous in thread | raw e-mail | index | archive | help
I'll take a stab at this based on real info and observation... > i 'll take the chance here, > > i've captured instances of netstat -m which reported 99% (even 100%) in use > and occasionally the machine crashed (and then i couldnt tell of course) and i was suspecting that "running out of mbufs" could well be the reason. > > At least that's what i liked to think since we are talking about "sligthly" modified TCP code with several connections opened simultaneoulsy with > large cwnds (around 17K each if i remember well). > > Am i right in assuming : > - "100% in use" will cause a crash ? I believe this will generate a call to the protocol drain routines. This shouldn't crash your box. It may freeze/pause network activity for a few seconds (up to 7ish on mine) until it's done with its thing. (This is based solely on observation.) 100% in use is also assuming that we're talking about 100% of max in use - which is probably the only condition you're likely to see this number anyway. > - the way to increase memory allocated to network is > options "NMBCLUSTERS=XXXX" ? Yes, so long as this number is greater than: #define NMBCLUSTERS (512 + MAXUSERS * 16) Otherwise it could decrease the number of clusters as was pointed out. To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-net" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199805080959.JAA12235>