Date: Sun, 08 Oct 2006 12:50:33 +0200 From: "O. Hartmann" <ohartman@mail.uni-mainz.de> To: Andrew Reilly <andrew-freebsd@areilly.bpc-users.org> Cc: "O. Hartmann" <ohartman@uni-mainz.de>, freebsd-stable@freebsd.org, Kris Kennaway <kris@obsecurity.org> Subject: Re: em, bge, network problems survey. Message-ID: <4528D7F9.6070807@mail.uni-mainz.de> In-Reply-To: <20061008043006.GA67655@duncan.reilly.home> References: <45244053.6030706@samsco.org> <20061005200552.GA80162@xor.obsecurity.org> <4525766D.5060004@mail.uni-mainz.de> <20061005214505.GA81829@xor.obsecurity.org> <4526201C.7060304@uni-mainz.de> <20061006172642.GA15047@xor.obsecurity.org> <20061008043006.GA67655@duncan.reilly.home>
next in thread | previous in thread | raw e-mail | index | archive | help
Andrew Reilly wrote: > On Fri, Oct 06, 2006 at 01:26:42PM -0400, Kris Kennaway wrote: > >> On Fri, Oct 06, 2006 at 11:21:32AM +0200, O. Hartmann wrote: >> >> >>>>> And so looks mine, FreeBSD 6.2-PRE/AMD64, >>>>> high I/O on disks and I/O on net renders this box unusuable ... >>>>> >>>>> >>>>> thor# vmstat -i >>>>> interrupt total rate >>>>> irq1: atkbd0 12437 1 >>>>> irq6: fdc0 27 0 >>>>> irq12: psm0 335285 42 >>>>> irq14: ata0 215 0 >>>>> irq17: fwohci0 1 0 >>>>> irq20: atapci1 102616 12 >>>>> irq21: ohci0+ 2 0 >>>>> irq22: nve0 ehci0 7594338 956 >>>>> irq23: pcm0 41007 5 >>>>> cpu0: timer 31752206 3999 >>>>> Total 39838134 5018 >>>>> >>>> You don't appear to be using the em driver. Can you confirm? >>>> >>>> Kris >>>> >>> positive. >>> >> It's probably a nve driver bug then, you should talk to the driver >> author. This may not be fixable in FreeBSD since it's a binary >> driver, so bug fixes mostly need to be done by the vendor. >> > > The OP may want to try the nfe driver, which I've been using very > happily on my amd64/nforce4 system for several months. > > I did for a while (most recent driver I could get from the mentioned japanese location) but the box got stuck after a while of network I/O. On the other hand, at this moment I play around with FreeBSD 6.2-PRE and do nearly every second day a build world, so it bothers me a bit doing the recommended patches. But your're wight. nfe() triggers a lot less interrupts than this binary Linux crap nve - for the time it runs. I would appreciate a soon occurence of the nfe driver in the STABLE or even in the 6.2-RELEASE code.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4528D7F9.6070807>