Date: Sat, 03 Jul 1999 12:20:35 -0400 From: Dennis <dennis@etinc.com> To: Stan Shkolnyy <stan@laurent.osgroup.com> Cc: hackers@FreeBSD.ORG Subject: Re: mbufs question/problem Message-ID: <199907031724.NAA17426@etinc.com> In-Reply-To: <Pine.LNX.3.95.990702160537.4189A-100000@laurent.osgroup.co m> References: <199906302150.RAA01512@etinc.com>
next in thread | previous in thread | raw e-mail | index | archive | help
At 04:10 PM 7/2/99 -0500, Stan Shkolnyy wrote: >On Wed, 30 Jun 1999, Dennis wrote: > >> >> I have a customer who has been experiencing "slow downs" with a freebsd >> router....they have substantially increased performance by reducing >> MINCLSIZE. I havent tracked the source, but im trying to hypothesize what >> it might be. On the surface I cant see any relationship since very few >> routines seem dependent on that value (m_devget() in particular, but I dont >> believe they are using any driver that use it). Is it possible that they >> are running out of small mbufs (they have NMBCLUSTERS set to a very high >> value)? >> >> Any ideas would be helpful. > >I have not noticed answers so far, so maybe their drivers copy mbufs very >often. AFAIK, "small" mbufs are indeed copied but "cluster" ones are not, so >when they forced the system to use more "cluster" mbufs, they got >substantial savings on copy operations. Well they are using Intel cards and our sync boards (of course our driver is binary so the change wouldnt effect our driver), and I dont see the fxp driver using small buffers. They are running bgp4...but I dont know what kind of buffers are used for routes. Dennis > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199907031724.NAA17426>