Comments on: Maximum theoretical downtime for a website: 30 minutes http://visualwebsiteoptimizer.com/split-testing-blog/maximum-theoretical-downtime-for-a-website-30-minutes/ Just another WordPress weblog Fri, 10 Feb 2012 10:32:09 +0000 http://wordpress.org/?v=2.9 hourly 1 By: Service Desk http://visualwebsiteoptimizer.com/split-testing-blog/maximum-theoretical-downtime-for-a-website-30-minutes/comment-page-1/#comment-713 Service Desk Fri, 23 Jul 2010 06:35:05 +0000 http://visualwebsiteoptimizer.com/split-testing-blog/?p=503#comment-713 Thanks for posting this, I better show this to my system admin so we can also study our website and network. Thanks for posting this, I better show this to my system admin so we can also study our website and network.

]]> By: Sean Reifschneider http://visualwebsiteoptimizer.com/split-testing-blog/maximum-theoretical-downtime-for-a-website-30-minutes/comment-page-1/#comment-537 Sean Reifschneider Tue, 22 Jun 2010 19:48:00 +0000 http://visualwebsiteoptimizer.com/split-testing-blog/?p=503#comment-537 At tummy.com, we do a lot of Linux high-availability work. To answer your question about whether the two servers in an IP fail-over situation need to be on the same switch: No, they don't. IP fail-over typically is done via sending a gratuitous ARP out when the IP change is done, so that things on that segment get the new MAC address for the IP. The MAC on each system stays the same, so a layer-2 switch doesn't see any change -- each machine's MAC stays on the same interface as before. It's the equipment on that segment (most importantly, usually, the router) that notice the change. They get the gratuitous ARP message and update their ARP tables, and start sending the traffic for that IP to the new MAC address. Once you start thinking about what is happening at the MAC level to pass packets around a local network segment, this all becomes pretty clear. Sean At tummy.com, we do a lot of Linux high-availability work. To answer your question about whether the two servers in an IP fail-over situation need to be on the same switch: No, they don’t.

IP fail-over typically is done via sending a gratuitous ARP out when the IP change is done, so that things on that segment get the new MAC address for the IP. The MAC on each system stays the same, so a layer-2 switch doesn’t see any change — each machine’s MAC stays on the same interface as before.

It’s the equipment on that segment (most importantly, usually, the router) that notice the change. They get the gratuitous ARP message and update their ARP tables, and start sending the traffic for that IP to the new MAC address.

Once you start thinking about what is happening at the MAC level to pass packets around a local network segment, this all becomes pretty clear.

Sean

]]>