Chris Menegay's WebLog

TFS scale-out

I got a little bit of clarification on scaling out Team Foundation Server on the chat this morning. Basically, it sounds like they just aren't planning for it or testing it. This seems odd to me for what amounts to an ASP.NET application being written to support hundreds of users. How does scaling out get pushed to a later release??? :-/

Obviously customers will end up testing this scenario for them, and as long as they didn't do anything violating MS's own best practices for web development, it should work. If it doesn't, I would expect MS to get verbally abused pretty bad for whatever design decisions prevent it.

Posted: Jan 27 2005, 12:18 PM by cmenegay | with 2 comment(s)

Comments

Paul D. Murphy said:

I just don't see how if you put the app tier on a dual proc XEON with a 3-4 gigs of ram you would need to scale it out.

It seems moot, IYAM.
# January 27, 2005 11:53 AM

Chris Menegay said:

The question was asked of me by someone at large consulting firm with thousands of developers. If they want just one TFS, it will probably need to scale. I agree that most companies shouldn't be able to overwhelm it, but large consulting firms, as well hosting providers (if they end up being allowed to do that) will be able to stress it. That's assuming it supports even 500 developers, as they've been saying.
# January 27, 2005 11:59 AM

BillE (Microsoft) said:

We did choose to go with the simplicity of a scale-up solution for V1 over scale-out. Part of the motivation for this decision had to do with the additional complexity that scale-out brings to deployment and administration of the tools and part to the additional testing burden needed to properly support it. This was a scoping decision that we made in the discussion that led to a 500 user team size. There are other features that complete the package when you consider the needs of teams of thousands. We arent building those features in this release.

The Visual Studio Team Foundation server is not far from being able to scale out in the MT, but we do have some disposable in-memory caches that are in play to help with performance. That state will thwart efforts to scale using NLB solutions. As a result, we are not looking for customers to test this out and let us know how it works. We know scaling out the MT won't work right now, but are trying hard not limit our options when we can build a complementary set of features to support teams of thousands.
# January 31, 2005 10:04 PM