March 1, 2006 at 2:53 pm
Microsoft.com is one of the most heavily used web sites in the world. And they don't cluster their database servers.
Wow! That is quite an admission on the Microsoft.com Operations blog. However there is a great explanation of why and that it doesn't fit the problem they are trying to solve. It's interesting reading, especially for anyone that is looking for a highly available web site.
It's also an interesting look at someone re-examining technology, and deciding what problem they needed to solve instead of just implementing the obvious choice. With so much read only information, it makes sense that a cluster isn't necessary. However their choice for the updateable information is an interesting one. It's a tradeoff between information loss and one that I think is worth making in many cases.
It's also fascinating to see that they put SQL Server behind a Network Load Balancing (NLB) cluster. That's something I've played with before, but the updates have always caused issues. But they're a great idea for read only stuff, if you can tolerate a few seconds delay when they are out of synch.
Plus there's always the 2516 user databases. That beats Andy's 250!!
Steve Jones
March 2, 2006 at 7:39 am
And to think i thought they ran microsoft.com on an Origami http://images.google.com/imgres?imgurl=http://www.gameshout.com/news/022006/images/022006_3361.jpg&imgrefurl=http://www.gameshout.com/news/022006/article3361.htm&h=150&w=200&sz=7&tbnid=Aqe82ZmdLaALUM:&tbnh=74&tbnw=99&hl=en&start=4&prev=/images%3Fq%3Dnews%2Borigami%2BMS%26hl%3Den%26lr%3D%26safe%3Doff%26rls%3DGGLG,GGLG:2005-34,GGLG:en%26sa%3DN
And an access 97 backend.
March 2, 2006 at 9:26 am
If you've ever worked with trying to cluster SQL Server, it's obvious why they didn't do it that way. It doesn't work.
March 2, 2006 at 9:32 am
I agree clustering is problematic. The infrastructure needed and stability of all the boxes needs to be sure... and the repair should shit hit the fane is IMHO having another pair or more of exact boxes lying around waiting for the time when it fails.
I guess I'm of the opinion in a mission critical situation when a catastophy occurs can you be really sure that mobo is not disaffected. Call it a healthy paranoia.
March 2, 2006 at 1:19 pm
2516 that's it! We have 3,229 databases.
March 2, 2006 at 2:00 pm
Couldn't you ave rounded it to a base 8 or 16 for better packet transmission, backup routines and compression ratios.
March 2, 2006 at 2:32 pm
Wow, that's a lot of databases. How many DBAs?
March 2, 2006 at 3:01 pm
Oh, yes it is! Currently most everything is automated to a point, so it's too bad to maintain, but it's currently just myself maintaining.
We store 17 months worth of data, which we currently create 172 new databases (and roll off about that many) each month. The total data for each month is around 400 GB.
Viewing 8 posts - 1 through 7 (of 7 total)
You must be logged in to reply to this topic. Login to reply