For the most part, I think it is fair to say "number of cores * time elapsed".
There exist a few caveats to that: 1) If you have a mixture of CPU speeds, you should probably normalize based upon the speed of CPU's. 2) If you have a lot of cores, each core has a bit less "oomph" than the previous one. That is, 72 CPU's ine one 15k domain is less powerful than 18 V440's. 3) Different platforms have hugely different costs. A 1 board 15k domain costs a lot more than a V440. Adjust your cost accordingly. 4) Not all applications are CPU bound. You may find that some run out of memory much sooner than they run out of CPU. In particular, I find myself buying more CPU's for J2EE servers so that I have more memory controllers. A lot depends on what you want to charge for. Are you charging based upon how much CPU time is used or how much benefit the consumer gets? If you are charging based upon benefit, you may want to beg Sun for the m-values for the systems you care about. Then, you could charge based upon the percent utilization (vmstat %usr + %sys in the simplest case) times the m-value of the system times the cost per "mvalue second" that is appropriate for the system. This message posted from opensolaris.org _______________________________________________ opensolaris-code mailing list [email protected] https://opensolaris.org:444/mailman/listinfo/opensolaris-code
