Date: Sat, 21 Feb 1998 14:08:10 +1030 From: Greg Lehey <grog@lemis.com> To: Bernie Doehner <bad@uhf.wireless.net>, "Alex G. Bulushev" <bag@sinbin.demos.su> Cc: Gabor Dolla <agdolla@datanet.hu>, freebsd-isp@FreeBSD.ORG Subject: Re: fault tolerant :)) setup Message-ID: <19980221140810.12820@freebie.lemis.com> In-Reply-To: <Pine.BSF.3.96.980220092225.5147A-100000@uhf.wireless.net>; from Bernie Doehner on Fri, Feb 20, 1998 at 09:25:06AM -0600 References: <199802200831.LAA13139@sinbin.demos.su> <Pine.BSF.3.96.980220092225.5147A-100000@uhf.wireless.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, 20 February 1998 at 9:25:06 -0600, Bernie Doehner wrote: >> infortrand RAID's allow fault-tolerant scheme with two or more servers, >> but it is a problem to mount the same UFS from different servers ... >> >> u can mount disk RW only for one server, after first server fail, >> the second server run fsck and mount disk (or mount disk RO) ... >> it is not good, but we use this scheme ... >> >> may be JFS :) > > What do you use on the backup sever to determine that the primary is dead? > I assume something like pinging the other host? That's the wrong way round. Pinging just tells you that the host isn't completely dead, not that it's up and running. What you need is for the system to tell you that it's still running. This is the method that Tandem uses: they're called "I'm alive" messages. > Is there any way you know of to set up a redudant system such that > the ISA cards from the primary sever are automaticaly hot switched > upon dead of the primary over to the secondary? Plenty, but they're complicated. The basic thing to think about is: any one component must be able to fail without compromising the integrity of the overall system. In our kind of environment, the most likely thing to fail is the network connection. That means you *must have* two network connections between the machines. If you don't, a network failure will make both machines thing that the other one has died. The resultant mess, which we called the "split brain syndrome" at Tandem, is horrible to behold. They don't both have to be Ethernet--in fact, there's merit in having, say, a parallel port connection and a network connection, or in this particular situation (sharing as SCSI bus), the SCSI bus itself. Greg To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-isp" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19980221140810.12820>