Remember that I was responding to the statement that it was impossible to have an IP address on two machines. ANYCAST is a way to do so. It's not the appropriate solution for most problems, but there are some cases where it works extremely well. I was in no way trying to say that ANYCAST and CLUSTERIP can be used for the same problems. They usually cannot.
An example of this that I heard of a few years ago is that Google uses ANYCAST on their corporate network. They have a HA pair of systems in each building that listen on a set of addresses that have been designated internally to be ANYCAST addresses and redirect all traffic directed at those addresses to a local set of servers. This allows Google to advertise a single set of IP addresses for common services and have those services be handled local to the user. You could try to do this by having many different DNS zones and serving the 'right' zone to the requester depending what their IP address is, but given that many clients will cache DNS lookups and you may move around fast enough (especially on a large campus) for this to be a problem, the ANYCAST solution is a very good fit. David Lang ________________________________________ From: [email protected] [[email protected]] on behalf of Dimitri Maziuk [[email protected]] Sent: Tuesday, August 21, 2012 10:05 AM To: [email protected] Subject: Re: [Linux-HA] IP Clone On 08/20/2012 06:01 PM, David Lang wrote: > ANYCAST has severe limitations on what you can do with it, but CLUSTERIP is > far > more flexible and can work in just about any local active/active problem. Apples have severe limitations on the amount of orange juice you can squeeze out of them, but oranges are far more juicy. -- in other words, that is misleading at best. Anycast is a router hack so it works over *routed* networks. Clusterip is *link-layer* broadcast so it works on single ethernet segment. One is for keeping core dns servers operational if the Internet breaks, the other is for when ldirectord is "too hard". One is for when multiple servers won't all reply at once because only one of them is visible to the reachable network, the other has a fixed rule that decides which server answers which clients. And so on. -- Dimitri Maziuk Programmer/sysadmin BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
