Date: Tue, 2 Aug 2005 21:44:00 -0700 From: Brooks Davis <brooks@one-eyed-alien.net> To: Dave+Seddon <dave-sender-1932b5@seddon.ca> Cc: freebsd-net@freebsd.org Subject: Re: running out of mbufs? Message-ID: <20050803044400.GA6079@odin.ac.hmc.edu> In-Reply-To: <1123040973.95445.TMDA@seddon.ca> References: <1123040973.95445.TMDA@seddon.ca>
next in thread | previous in thread | raw e-mail | index | archive | help
--SUOF0GtieIMvvwua Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Wed, Aug 03, 2005 at 01:49:32PM +1000, Dave+Seddon wrote: > Greetings,=20 >=20 > I'm trying to do some performance testing of a content filtering system, = so=20 > I'm trying to get very high HTTP throughput. I've got 4 * HP DL380s with= =20 > 3.4G Xeon processors (hyper threading) and 1 G RAM, 2 onboard BGEs, and 2= *=20 > 2 port EM. Using FreeBSD5.4-stable (as of 2005/08/02) and device polling= ,=20 > I've configured a large number (246) VLAN interfaces on two machines, and= =20 > have apache on one box and siege on the other. Using 'siege -f=20 > /home/my_big_list_of_urls -c 50 --internet' one host does a large number = of=20 > request from the other machine. I've been trying to tune for maximum=20 > performance and have been using lots of examples for /etc/sysctl.conf and= =20 > so on from the web. Adjusting these settings and running the siege, I've= =20 > found the apache server completely loses network connectivity when device= =20 > polling is enabled. I've adjusted the HZ lots and found the system=20 > survives the longest set a 15000 (yes it seems very large doesn't it). T= he=20 > problem now seems to be that I'm running out of mbufs:=20 I doubt this is relevant, but it's worth noting that the vlan code is is rather stupid in the face of large number of vlans because they are all hung off a last. Thus assuming even load, you're traversing an average of 123 elements in the list per packet. This might have a significant impact on performance that could be causing excessive queuing. It's fairly easy to fix by allocating an array indexed by vlan so dispatch is constant time (modulo cache effects), but it hasn't been done yet. -- Brooks --=20 Any statement of the form "X is the one, true Y" is FALSE. PGP fingerprint 655D 519C 26A7 82E7 2529 9BF0 5D8E 8BE9 F238 1AD4 --SUOF0GtieIMvvwua Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.1 (GNU/Linux) iD8DBQFC8EuPXY6L6fI4GtQRAnFnAKC19kpGJubV7ztj+HUDXw1dpCEoDgCfQI9o HNUslpupZv8ypAm/R5M5Zxo= =Go9u -----END PGP SIGNATURE----- --SUOF0GtieIMvvwua--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050803044400.GA6079>