MSCOM WebCast Week…Q & A (Part 1) High Availability Architecture with Microsoft.com Operations
First of all thanks so much to everyone that was able to tune into the Webcasts series that we finished up on Friday. We had a blast doing these and we hope that you got some good information out of them. We certainly got some good questions from you.
I would like to give some kudos to Chris Adams from the IIS product team who served as the Master of Ceremonies for us this week. Chris is a member in good standing of the blogosphere, (as well as an all around good guy) and has a wealth of information about IIS, so check out his blog as well at http://blogs.technet.com/chrisad.
If you missed these webcast and want to see them “on demand” I will include the links so that you can access these and watch them at your leisure. First things first, and that would be Monday’s kickoff session on High Availability Architecture. Chad Kraykovic and Brad LeRoss covered our availability goals with key information about our partnerships; how we measure availability; then they dug deep into the high availability architecture that included: software and hardware platforms, host level load balancing, global solutions and networking, web and database server distribution and database system architecture.
Monday’s topic was: High Availability Architecture with Microsoft.com Operations
For an on demand viewing of this webcast go to:
http://msevents.microsoft.com/CUI/EventDetail.aspx?EventID=1032283678&Culture=en-US
Here is the recap of the questions:
Just 12 application pools, I would think you would utilize more pools for application protection through process isolation, no?
The web applications are grouped in “trusted” or “non-trusted” AppPools. It will be covered in more detail later in this presentation.
So is NLB the only load balancing solution utilized? No hardware load balancers in the architecture at all?
Yes we use NLB for local network load balancing on Microsoft.com. We find this to be the simplest solution that enables us to use a software solution to monitor and react to servers having issues in the environment. When a server is not responding correctly we can then easily take the server out of rotation. We have determined that the sweet spot for the number of nodes in a cluster is eight in our environment. We have a different solution for Global Load Balancing.
For these non-microsoft services you mention, how can we find out more information about costing, etc.?
The best approach will be to directly contact the third party service provider. We currently work with Akamai (http://www.akamai.com) and Savvis (http://www.savvis.net).
If log-shipping is being used to "synchronize" primary and secondary servers, what interval of time is used between shipping? Is the secondary server completely inactive while the primary server is active?
We set replication intervals based on application requirements but usually this is set at every 5 minutes when used within this solution. The secondary server is offline and if the first server goes offline we would have to manually switch to the secondary.
Why not use DNS aliases to facilitate quick role change , rather then webconfig change ?
The web.config change gives us the ability to redirect traffic faster. There is TTL on the DNS record as well as client side DNS caching compared with almost instantaneous change when web.config is updated.
What is the overhead to Performance on the server during the shipping of logs (i.e. Log Shipping)?
The performance cost of replication changes drastically based on the amount of data that needs to be replicated at each interval and any code that is being run on the server being replicated to. This being the case performance costs of replication can only measured on an application by application level.
Will database mirroring be expected to work across multiple datacenters? If so, what minimal bandwidth might be expected between datacenters to make this work?
Yes, mirroring will work across datacenters. However, to utilize mutiple datacenters you'll want to run in asynchronous mode, which means there can be latency between the priciple and mirror server. Bandwidth will depend on the volume of changes on the principle server. Unfortunately the answer is to test, test, test.
Is there some kind of implementation guide for WOW64, to assist in making 32-bit applications to successfully be migrated to 64-bit Windows? Also, was there a compelling business case on this before it was done?
There is a great Knowledge Base Article KB895976 How to run 32 bit worker process on 64 bit.
So do you need a x64 bit version of scripts as well?
You do for the scripting host, in other words the 64 bit version of cscript or jscript. Look at the next issue of Technet Magazine in the Inside Microsoft.com column, the entire article is about MSCOM adoption of x64 bit for www.microsoft.com, it has a pretty complete story.
