Performance for D0 Nodes

When you want to run a job on the D0 network you need to know the relative performance of the node that you choose to run your job on. Here is a table with some of the public benchmarks on it. If you can help me fill in the blanks send me email at rhilgers@fnal.gov. Note all of these figures are basically meaning less. Each benchmark tests things that you will probably never use and they can vary depending on operating system and compile options and the efficiency of the compiler or the processor its self. It is possible to engineer a cpu that performs very well on these benchmarks and is a poor performer in everyday use.

I also put some Intel chips in for comparison only. Your results may vary on an intel chip depending on the operating sytem and compiler. In my benchmarking and testing Dos will beat windows, windows 95 or windows NT any day. This is because DOS only runs one process at a time so the CPU runs the job 100% of the time. All of the other operating systems run several processes at the same time and unix runs the most but also does the best job multitasking.

Use this performance chart along with the system status report to find a system to run your programs on. Also if you know your code is more integer intensive you may choose a different system to run it. If your job is heavy in graphics then you will want a system that has a fast graphics processor like d0sgi9, d0sgiucr2 and d0sgifsu1, however, you will need to be working from those workstations in order to take advantage of the graphics processor..