Stress Test Blades & Blade standardization with applications

Thanks for posting ! This really depends. The best tool to stress the blades is the real application that you will be using for production. That's the only true way to get a "feel" for if they will work for your particular application.

Blades, especially now that we have so many more choices with respect to memory and processors, are a good fit for many applications. I've got two M905's in a chassis right now with 4x quad core procs and 96GB or RAM. That's some serious horsepower.

With plenty of IO connectivity in terms of SAN and Network, not much you can't throw at it. It's nice that we have all finally agreed that the data drives for servers really need to be on a SAN, either iSCSI or Fiber, and blades fit well in both.

Internally we use the standard Exchange stress test tools, and the DVD Store ( http://www.delltechcenter.com/page/DVD+Store) written in house to run tests on the servers we have in the lab. These are publicly available tools that you can use also.

With servers there is not a "slap it on and getting in running quick" benchmark, like you have with PC's. So many more variables in what you need to measure. Some people try to do xcopies or linux cp commands to measure IO throughput, but that is the WRONG way to go. When doing this, people are running single threaded IO commands against servers and storage that were designed for 100's 1000's are more users hitting them at the same time. They are measuring the wrong metric.

It can be explained in this analogy, which a very smart person in a performance lab told me long ago ........

Which is faster ? .. an F1 race car, or a school bus ? .. depends, is the task at hand taking 1 child to school or 100 ?

Servers are better for *real* workloads, PC's will almost always win a *copy* contest. I'll get someone else to weigh in on this thread as well.