Date: Sat, 15 May 1999 13:22:44 -0700 From: Graeme Tait <graeme@echidna.com> To: freebsd-isp@freebsd.org Cc: info@boatbooks.com Subject: Redundant servers Message-ID: <373DD794.15EF@echidna.com>
next in thread | raw e-mail | index | archive | help
I operate a colocated server (busy WWW, plus primary DNS and a small amount of mail) for some clients, and we have been thinking about how to assure (almost) uninterrupted service in the event of server failure. The problem is that if physical access is required (e.g., because of hardware failure), we may at times have difficulty achieving an acceptable repair time. There is also the issue of minimizing downtime during major upgrades. Since this machine takes online orders, downtime is directly translatable into dollars of income lost. The existing machine is a plain PII-400 with 512MB RAM and 18GB total non-RAID SCSI disk, and has been perfectly reliable to date. We could purchase a fancy "server-grade" machine with RAID, redundant power supplies, etc., but the fact is that there are still things that could fail, and we'd still have to go down for certain upgrades. Also, such a machine would be so expensive that having a spare would be out of the question at present. For less total cost, we can commit a machine identical to the existing server as a hot-standby, installed in the colo. So my first question is does anyone have opinions on this or alternative strategies for maximizing uptime, given that physical access to the server may be severely restricted? The second issue is how to actually run a hot standby, and effect a switch on failure of the primary. I would want the spare machine on the network, since it could be used for offloading some background tasks from the live machine, and also for backing up some critical data. My thought was to assign the spare a separate primary IP address, and have only that address active under normal circumstances. If it became necessary to switch over from the live server, we would somehow take the live server off the network (by software, remote power-down or physical disconnection), and then run a script on the spare to ifconfig the 50 or so IP alias addresses that the operational server answers to for DNS, mail and WWW traffic. Maybe this could be automated based on monitoring the primary machine. The other issue is how to make sure the spare machine is basically a mirror of the primary machine. -- Graeme Tait - Echidna 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?373DD794.15EF>