Switching generators from small client towers to servers?
We currently have 12 load generators that are used for Loadrunner. They are all Intel Core i5 2500k CPUs, 4GB of RAM, 250GB HDDs. During the biggest tests that these generators run, CPU usage is around 80% across them all. RAM isn't hit terribly hard, though. They are separated into three banks of 4, and each bank has their own switch that has a 1Gb uplink (the switches themselves are also 1Gb switches). I'm considering switching over to some higher powered blade servers. I'm hoping to build a couple higher powered servers to take the place of these 12 generators. Each server would be in different server rooms so they will have separate Ethernet connections. Each server will have a 1Gb connection, though we could possibly do better than that.
Has anyone had any experience with this? I'm hoping to do this so I have less machines to maintain as well as the space it's going to free up.
Your approach sounds totally reasonable to me. We are constantly upgrading our Load Farm to higher capacity servers and have seen huge increases over the years in terms of how many vusers we can run per LG. I'm guessing that your approach will yield similar results. Moving into "server class" machines should make a big difference! You might consider standing up one server in your proposed configuration and then doing a few baseline tests to see how many servers you may ultimately need.
SoCalGal - Defender of end user response times!
With LR 12 the switch has finally been made to 64 bit load generation. This effectively uncaps the 4GB limit for virtual user code. As to whether to go with Blades or VM's, I would hold to a core group of blades and use those for your high quality samples and then examine the differences in your VM environment. There are several core challenges with VMs. One core item is the floating clock in the VM, which HP has effectively put to rest on VMWARE environments with the right patches and load generators settings enabled. Other items which are more difficult are the trend towards placing Load Generators on already overloaded VM hosts which causes all sorts of initial condition and in-test repeat-ability issues.