Date: Fri, 13 Aug 2004 19:47:40 +0300 (EEST) From: igor@ktts.kharkov.ua To: oberman@es.net (Kevin Oberman) Cc: freebsd-stable@freebsd.org Subject: Re: 5.2.1 performance issue Message-ID: <200408131647.i7DGlecR016845@ktts.kharkov.ua> In-Reply-To: <20040813160611.74F845D04@ptavv.es.net> from "Kevin Oberman" at Aug 13, 2004 09:06:11 AM
next in thread | previous in thread | raw e-mail | index | archive | help
The problem is neither duplex nor WITNESS/INVARIANTS. And as I outlined before there is no switch, but a router. Connection is 1000/full. I checked everything I could 1000 times. I started believing that 5.2 performance of 5.2 is much worse than 5.1. Does anybody here run servers with high speed connections and lots of TCP sessions? Did you ever experience something similar? Igor. In original message Kevin Oberman writes: > > > Date: Thu, 12 Aug 2004 19:19:09 -0700 (PDT) > > From: Doug White <dwhite@gumbysoft.com> > > Sender: owner-freebsd-stable@freebsd.org > > > > On Wed, 11 Aug 2004 igor@ktts.kharkov.ua wrote: > > > > > I noticed that box started getting input errors on em0. > > > In order to fix that I increased EM_MAX_TXD from 256 to 1024 and > > > EM_MAX_RXD from 256 to 2048. > > > After that Input errors disappeared but I continued getting intr_queue_drops > > > counter increasing. > > > > You might double check your duplex settings, and if you're on a managed > > switch, the stats there. It may not have autonegotaited properly after > > the reboot. > > > > Also make sure you aren't running with WITNESS and INVARIANTS. Those are > > big CPU pigs. > > WITNESS is certainly a big CPU hog, but I don't see why INVARIANTS would > be all that significant. And both are disabled in 5.2.1 GENERIC. > > I'd bet on either a duplex mis-match or an auto-negotiation > failure. Both are very common and the latter can be a pain to deal > with. (Don't assume that you can just nail up one end.) > -- > R. Kevin Oberman, Network Engineer > Energy Sciences Network (ESnet) > Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab) > E-mail: oberman@es.net Phone: +1 510 486-8634 >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200408131647.i7DGlecR016845>