How well does Anycast work with Windows 2000 or XP servers? Is the Microsoft OSPF implementation good enough to use or do people port another routing implementation? Yeah, I know about Unix/Linux. All the large scale anycast deployments I know about are on unix, but I was wondering if anyone has tried it using windows.
On 18 Feb 2004, at 22:43, Sean Donelan wrote:
How well does Anycast work with Windows 2000 or XP servers? Is the Microsoft OSPF implementation good enough to use or do people port another routing implementation?
Does the capitalisation of Anycast mean it has been productised, now? :-) In my head, anycast is unicast without globally-unique addresses. Since nobody ever asks "does unicast work with Windows?" I am guessing you mean something different by "Anycast" than I do by "anycast". What do you mean? You can do the kind of Internet-wide service distribution that we do with F [1] with no unusual requirements on the host. (ISC in fact deploys clusters of hosts in each node rather than a single host, and the way we build the clusters involves OSPF on the hosts, but that's orthogonal to the wide-area distribution strategy.) Joe [1] ISC-TN-2003-1, http://www.isc.org/tn/
On Feb 18, 2004, at 10:43 PM, Sean Donelan wrote:
How well does Anycast work with Windows 2000 or XP servers? Is the Microsoft OSPF implementation good enough to use or do people port another routing implementation?
Yeah, I know about Unix/Linux. All the large scale anycast deployments I know about are on unix, but I was wondering if anyone has tried it using windows.
Honestly, I do not know about OSPF (or BGP) on Windows, however, you can just static route to the Windows box(es). Sure, if the OS hangs, the interface will stay up and the static route will still push bits at the dead box, but it will work (FSVO "work"). Besides, how often does Windows crash? <snicker> -- TTFN, patrick
On Thu, 19 Feb 2004, Patrick W.Gilmore wrote:
Honestly, I do not know about OSPF (or BGP) on Windows, however, you can just static route to the Windows box(es). Sure, if the OS hangs, the interface will stay up and the static route will still push bits at the dead box, but it will work (FSVO "work").
Besides, how often does Windows crash? <snicker>
Hence the reason why I want the route to cease being advertised if the box "fails." I'm trying to avoid putting yet another server load balancer box in front of the windows box to withdraw the route so a different "working" box will be closest. It may be an oxymoron, but I'm trying to make the windows service (if not a particular windows box) as "reliable" as possible without introducing more boxes than necessary.
Sean,
Hence the reason why I want the route to cease being advertised if the box "fails."
I'm trying to avoid putting yet another server load balancer box in front of the windows box to withdraw the route so a different "working" box will be closest. It may be an oxymoron, but I'm trying to make the windows service (if not a particular windows box) as "reliable" as possible without introducing more boxes than necessary.
You might be better not running the routing protocol on the Windows box, and run gated (or whatever) on some nearby Linux/BSD box which tests the availability of each of your windows box and introduces the appropriate route (i.e. a next-hop for the anycast address pointing at a normal IP address) into (a very local) BGP (running multipath) or other favorite routing protocol for each of the servers that are up. Alex
At 05:43 AM 2/20/2004, you wrote:
Hence the reason why I want the route to cease being advertised if the box "fails."
I'm trying to avoid putting yet another server load balancer box in front of the windows box to withdraw the route so a different "working" box will be closest. It may be an oxymoron, but I'm trying to make the windows service (if not a particular windows box) as "reliable" as possible without introducing more boxes than necessary.
You haven't said what type of service you want to make as reliable as possible. It sounds like you want to use clustering or network load balancing. With clustering, you can have the service present on both machines and if the link between the two fails or if the service on the primary machine fails, the second machine will take over. You can also use shared Fiber-channel or SCSI devices between the two servers. You can also use network load balancing to share a non-transaction based service between servers. If you do it this way, you will get automatic load balancing to double the speed and capacity between the two or more servers in the NLB cluster since they all service requests all the time. In both cases, you will create a virtual IP address which receives all connections and both machines in the cluster will determine which machine handles each connection. This isn't hard and we do it all the time. -Robert Tellurian Networks - The Ultimate Internet Connection http://www.tellurian.com | 888-TELLURIAN | 973-300-9211 "Good will, like a good name, is got by many actions, and lost by one." - Francis Jeffrey
At 05:43 AM 2/20/2004, you wrote:
On Thu, 19 Feb 2004, Patrick W.Gilmore wrote:
Honestly, I do not know about OSPF (or BGP) on Windows, however, you can just static route to the Windows box(es). Sure, if the OS hangs, the interface will stay up and the static route will still push bits at the dead box, but it will work (FSVO "work").
Besides, how often does Windows crash? <snicker>
Hence the reason why I want the route to cease being advertised if the box "fails."
Connect the server(s) to APC MasterSwitch or equivalent hardware. Monitor the server box(es) for responsiveness. If/when it fails, the monitoring station can instruct the MasterSwitch to reboot (power cycle, really) the box. Stuff is pretty inexpensive (certainly less so than load balancers).
I'm trying to avoid putting yet another server load balancer box in front of the windows box to withdraw the route so a different "working" box will be closest. It may be an oxymoron, but I'm trying to make the windows service (if not a particular windows box) as "reliable" as possible without introducing more boxes than necessary.
My initial thought last night was in fact the use of load balancers. But then you need to think about redundant load balancers and so on.
participants (6)
-
Alex Bligh -
Daniel Senie -
Joe Abley -
Patrick W.Gilmore -
Robert Boyle -
Sean Donelan