TFS Build errors TF224003, TF215085, TF215076

I am using TFS2008 and VS2008. I run nightly builds for about 20 applications using one build agent and the builds are scheduled for either 1am or 2am. Most of the build succeed, however 6 of them fail regularly with similar errors. The errors are either the first two below, or the third one by itself:

• TF215085: An error occurred while connecting to agent \xxxx\BUILDMACHINE: TF215076: Team Foundation Build on computer BUILDMACHINE (port 9191) is not responding. (Detail Message: The request was aborted: The operation has timed out.) 11/04/2010 2:10:10 AM
• TF224003: An exception occurred on the build computer BUILDMACHINE: The build (vstfs:///Build/Build/2632) has already completed and cannot be started again..
• TF215085: An error occurred while connecting to agent \yyyyy\BA_WKSTFSBUILD: Team Foundation services are not available from server srvtfs. Technical information (for administrator): The operation has timed out

It looks to me like some kind of communication error, maybe the port gets over 开发者_如何学Goloaded - can this happen? Should I spread the builds out a bit more? In the build definition it says "Queue the build on the default build agent at", so I figured if I scheduled them to start at the same time they would be queued and occur sequentially.

Most of the suggestions I've found online for these errors are for all or nothing scenarios where no builds work at all whereas my problem is most build but some consistently do not. Note that if I schedule the builds via Team Explorer -> BuildDefinition -> Rightclick - Queue new build they run without error. Judging by the dates of the last successful builds of these 6 failing builds I believe it is the same 6 failing every night. (I'm editing the build definitions now to keep the failed builds so I can get some more info on the problem)

It's been a while since I've managed builds with TFS 2008, but I believe I remembered running into memory issues if too much was run on the same server. There are some parallel flags you can use with MSBuild that may speed it up. I believe if there's too much recursion (too much solution nesting) it can cause some timeouts like this. You may also consider removing your agents and recreating them, especially if the same builds run in some agent workspaces but not others. It seems like you were saying some builds consistently fail, which in TFS 2008, I believe are mapped directly to dedicated agents. Try the same builds using different agents and see what happens.