| || |
Potential issues moving Load Generation to AWS cloud?
We are considering moving our physical load generation environment of ~30 local generators to the AWS cloud. Does anyone out there have any experience with a similar move and what "gotchas" we might need to look out for?
Thanks in advance!
The same rules which apply to local virtualization of load generators also apply to cloud virtualization.
You can't control your network conditions initially or in test. This leads to higher test to test variability.
You can't control what other people are going to be running on the same VM host on VMs you do not own manage or control, whose resource requests will be adjudicated by the same hypervisor which is making decision on your VM resource access. Sometimes it will benefit you, sometimes not.
Your clock will float inside of the virtual machine. You cannot control this. This will lead to higher averages, maximums and standard deviation values. Not because the app is slower but because the clock will jump while you have timing records open.
You pay for every resource hit: CPU Cycle, Disk Cycle, Network Cycle, Memory Cycle, on your load generators. Depending upon how you choose to tax your load generators and how long you leave them up the "bargain" disappears rapidly.
Hedge your bets. Leave half a dozen load generators inside as a "reference" or "control" set. Make your decisions based upon the metrics from this set of load generators, not the ones in the cloud.
Replace ineffective offshore contracts, LoadRunnerByTheHour
. Starting @ $19.95/hr USD.
Put us to the test, skilled expertise is less expensive than you might imagine.
Twitter: @LoadRunnerBTH @PerfBytes
Originally Posted by jpulley3
Thanks for the great feedback, hadn't considered the "floating clock" aspect of the cloud. I would this item #1 (shared VM resources) could be minimized by using "reserved" machines, but this leads to higher cost also. Great input!
Follow up question: Does anyone know if performance center can dynamically request virtual servers from the Amazon cloud, thus allowing for true "only when needed" controller / generator availability?