I have now finished running the cooltst.ksh script on my large simulation. Now, I am getting a Yellow result with a % of around 1.1 to 1.2% . How do I interpret this? If I will only have 1 process in use at a time, is it likely to matter if I get a borderline value?
Many thanks for your assistance.
According to the readme that accompanies cooltst, "The floating point analysis looks at the ratio of floating point instructions to the total number of instructions. If less than 1%, this will be marked green; between 1% and 3%, yellow; greater than 3%, red. Even a red indicator does not preclude a workload from being executed on a T1 processor; however, a red indication is a warning that the workload under test may strain the floating resources of the processor, particularly if other high floating point workloads are simultaneously executed. The T1 processor performs best when threads cover for other threads waiting on memory." The thresholds are based on rules of thumb derived from the Programmer's Reference Manual and experience. They are intended to draw attention to a potentially problematic area so that one proceeds with due caution.
You seem to suggest that your simulation process makes use of the FPU. You can verify this with cputrack, which is available under Solaris. You should be able to borrow the argurments used by cpustat and use them for cputrack. Obviously the FPU is not a potential bottleneck for processes that are not executing instructions that use the FPU.
I suspect that your workload would run on a T2000 server without the FPU being a bottleneck given your brief description. If you would send me the files in the directories that cooltst creates, I would be happy to look quickly at the cpustat summary file to confirm my suspicion that the FPU is not problematic.
Based on your description, I have some concern that your application environment is single threaded: most simulators are. T2000 servers are optimized for throughput computing and excel at executing parallel tasks. cooltst provides information about processes and their core utilization. You should look at the cooltst process/LWP output to verify that your application environment has opportunities to take advantage of the US-T1's parallelism and does not have a single thread of execution that can act as a bottleneck.
With the Try and Buy program, you can always empirically test your application environment and its runtime characteristics.
