As I mentioned earlier we were running scalability tests on a Unisys 8-way box as well as the various 1P - 4P configurations.  I didn't publish the results at the same time because they didn't make any sense.  The "big" 4 way config achieved something on the order of 220 RPS and was CPU bound.  Our initial 8-way test was only able to achieve about 280 RPS and was also CPU bound.  Given twice the processors, that should not have happened.

We did some investigation and found a problem with the indexes and with a stored procedure in our data warehouse.  These problems substantially limited high-end scalability.  After fixing them, we are now getting 381 RPS - much more inline with expectations.  Using the math below and rounding down a bit - I'd estimate this as supporting a team of about 3,500 people!  That's pretty fantastic in my book.  There aren't many development team in the world bigger than that.

So, the official results are:

Team Size

TFS config

Model

CPU

Memory

Disk

3,500

Dual Server

AT: Dell PowerEdge 1850

DT: Unisys ES7000/600

AT: 2P 2.8Ghz,

DT: 8P 2.8Ghz

AT: 4GB

DT: 32GB

XIOTech SAN - 2TB, RAID10, built with 140GB 10K RPM drives

I want to publicly thank both Unisys and XIOTech for working with us on these high end benchmarks.  They've been incredibly helpful working with us on hardware and configuration.  And I'm happy to be able to attest to the fact that we have seen some fantastic results with their products.

We're going to go back and re-run the 4-way numbers with the warehouse fixes and I do expect to see some improvement there but I'm not expecting it to be dramatic at that scale.

And just to add the ever important caveat - your mileage will vary.  I can't (and don't) promise that you will be able to replicate the results that we achieve.  There are many variables that affect performance.  You should use this only as a general guide for your deployments and do your own testing to ensure your hardware is properly sized for your needs.

Brian