Frank Brown ([log in to unmask]) wrote: : >>>>>>>>>>>>>>> : Regarding Vals note beginning ... : Can anyone help me in estimating TRUE perfomance capabilities of new : HP3000 : 959/100 - 959/300 machines? : ............... : I too am interested in the true performance of all of HP's current menu of : machines vs. older models such as 70 and our current 960. Anybody : have any 'real' insight? What do you want to know. The MPE V/e systems had slow CPUs but were efficent users of memory. The typical MPE v/e system was DISK bound with available CPU. Then along came Disk caching and some systems got a boost in I/O performance but still had the slow CPU. Our relative performance numbers are typically based on benchmarks and they generally are expressed in terms of throughput. For example a S995/100 get xx times more throughput than a S960. You do need to be aware of the differences if you are crossing the V/E to i/X bridge but the performance numbers are good indicators of relative performance. They DO NOT indicate the relative power of the CPUS. The PA-RISC CPUs actually have higher relative CPU performance differences than our benchmark relative performance numbers. A good counter example is the S955 to the S960. They were the SAME CPU but they had different relative performance because the larger caches(on the S960) help OLTP. In batchmode the larger cache did not help and we documented that batch performance might be the same on both machines. The relative performance of a S995 vs a S959 really depends on usage. The S995 has larger caches but is 90mhz instead of 100mhz. Of course we assume in our numbers you have a CPU bottleneck and faster CPUs will get more throughput. If you are I/O bound or have lock contention or some other kind of bottleneck then you will not get the same relative performance we publish. If the application already runs on MPE i/X then you could expect to get something like our relative performance numbers if you move to a faster box. The hardware/software combination of MPE V/E and MPE i/X is different enough that occasionally things don't exactly follow our relative performance numbers. The numbers Mike Yawn posted are reasonable indicators of relative performance(throughput) if you are currently on a i/X machine and you don't have any other bottlenecks. As a closing word, remember the old performance adage "it all depends". -- ************ __ *********************************************************** ********** / / *********** Stephen Macsisak : Performance Consultant * ******** / / ******** Commercial Systems Division (CSY Labs) * ****** / /___ ______ ****** arpa: [log in to unmask] * ***** / __ // __ / ***** uucp:(hplabs.sun.uunet)!cup.hp.com!stevemac * ***** / / / // /_/ / ***** hpdesk: Steve Macsisak /hp4700/m2 * ****** /_/ /_// ____/ ****** US Mail: 19111 Pruneridge Ave, MS 44U2 * ******* / / ******** Cupertino, CA 95014 * ********* / / ********** Phone/Telnet: 408-447-5851 / 1-447-5851 * ********** /_/ *************************************************************