Date: Thu, 23 May 1996 08:38:58 -0700 From: "Justin T. Gibbs" <gibbs@freefall.freebsd.org> To: Chris Linstruth <cjl@qnet.com> Cc: freebsd-hardware@freebsd.org, freebsd-isp@freebsd.org Subject: Re: FreeBSD: SCSI hangs, panics and other failures as nntp server Message-ID: <199605231538.IAA20647@freefall.freebsd.org> In-Reply-To: Your message of "Thu, 23 May 1996 06:48:00 PDT." <m0uMakP-0003ioC@cello.QNET.COM>
next in thread | previous in thread | raw e-mail | index | archive | help
>Dear FreeBSD-Hardware (and ISP): > >We at qnet have been attempting to use FreeBSD as a >Usenet News server for about a year. We are at this time >evaluating other alternatives because we have been unable to >achieve the stability required under this kind of load. > >We are perfectly satisfied with 2.1.0-RELEASE's performance as >a Domain Name Server, mail hub, and W3 (apache) server. > >We have built several different machines based on the AHA-1542, >NCR83C510, and the most recent being an AHA-2940 (AIC7870). Don't use 2.1.0R with a 2940. Use -stable. >We have been unable to build a system that does not suffer >from: > > SCSI Bus hangs - (you can switch VTs etc, but anything > accessing any SCSI device terminally hangs. Telnetting > in results in a connection but no login prompt, but > characters echo. You know what I'm talking about. > > Random Reboots > > Device Timeouts - with 2.1.0-RELEASE. We just upgraded > to the 2.2-960501-SNAP and now we're getting reboots > instead of system hangs. (detailed in kern/1157) If you are getting reboots, its most likely not a device timeout. You should probably be supping the CVS tree and relying on 2.1-STABLE not on current. >Is there anything we can do? We've replaced every component including >SCSI cables, terminators, etc. We've tried external SCSI cabinets, all >internal devices, and combinations of both. Try -stable. Its designed for your type of aplication where stability is more important than new features. >We would certainly like to continue using FreeBSD for all our >subsystems, including Usenet. Where should we look next? > >Would another host adapter (Buslogic?) solve all our problems? Perhaps. The Buslogic driver may be more stable, but offers poorer performance than the 2940 driver. There seem to be two things that can trip up a 2940 right now. Mixing wide and narrow devices, and turning on ultra mode. I think I have the ultra problem squashed, but I don't have any wide or ultra devices to test with. >If you were going to buy a new PCI host adapter for a FreeBSD system >what would it be? Probably a 3940UW. The aic7xxx driver is the most actively maintained driver in the source tree and the usual response time on bugs is good. >What hardware deficiency would typically cause these symptoms? Poor termination, device driver bug, VM bug... Go back to -stable and see how that works for you. >Does the -stable kernel contain anything related to this that the 960501 >snapshot does not? -stable is a totally different source tree then what was in the 960501 snap. >Thanks in advance for *any* assistance. > >(A project for today includes recompiling a kernel with the appropriate >debug options available, so that info should be available soon.) Try to compile a custom kernel that has only the devices you use in it. -- Justin T. Gibbs =========================================== FreeBSD: Turning PCs into workstations ===========================================
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605231538.IAA20647>