OBERMAN@ICDC.LLNL.GOV ("Kevin Oberman, LLNL, 422-6955, L-156", 415) (06/16/88)
>Anyone had experience ( good/bad/indifferent ) of the RABBIT-9 package > alternatively known as VAST - VAX Acceleration Software Technology? > The package dynamically changes system parameters every two seconds > and on a VAX 8700 only uses about 0.2 sec CPU time per day ( as reported > by SHOW SYSTEM ) and, from my observations, appears to ensure that there > is always an adequate pool of free memory : adjusts FREELIM,etc dynamically! I tried both RABBIT-9 and DLB (a competitor) on nearly identical nodes in our cluster. Both seemed to work pretty well, but a comparison of SPM output seemed to imply that DLB was a LITTLE BIT better. The difference was within the relm of statistical error, so I could not pick either product based on performance. Both made a significant improvement on the CPU utilizatiuon of our rather memory rich 780 systems. Neither product had any negative impact on our system, but I did have one concern with RABBIT-9. I was monitoring SYSGEN parameters rather closely during the test and I noticed RABBIT-9 changed some parameters not mentioned in the documentation. (Or at least ones I couldn't find mention of.) While none of the changes were `bad', I was not happy to see them messing around with the system in ways not documented. A final concern was that RABBIT-9 lowerd QUANTUM to 2 and IOTA to 0. I felt that these were ab bit drastic and would probably increase overall system time. But, I can't say just what the significance would have been if the CPU load got REALLY bad. DLB made less radical changes in the system with very similar results. Its monitoring tool was far less well done than RABBIT's excellent one. While both systems can be biased to favor interactive over compute bound jobs, the more radical approach of RABBIT might be a bit more effective if fast CPU response is critical. The cost would be somewhat higher system overhead. I'm sure both vendors will continue to tweek their products and the results may very well converge. The final consideration is pricing policy. RABBIT-9 is priced according to the processor running it. A uVAX is cheap and an 8700 is expensive. DLB is a flat price regardless of the processor. Since both seem to work pretty well, I think that I might let the price be a determining factor if you are not on a 78x system. (They are the same price in 78x systems.) R. Kevin Oberman Lawrence Livermore National Laboratory Internet: oberman@icdc.llnl.gov (415) 422-6955 Disclaimer: Don't take this too seriously. I just like to improve my typing and probably don't really know anything useful about anything. And my boss is not in any way recommending anything. These comments and opinions are my own.
frei@rubmez.UUCP (Matthias Frei ) (06/30/88)
In article <8806281702.AA25188@ucbvax.Berkeley.EDU>, Kevin Oberman writes: > A final concern was that RABBIT-9 lowerd QUANTUM to 2 and IOTA to 0. I felt > that these were ab bit drastic and would probably increase overall system > time. But, I can't say just what the significance would have been if the > CPU load got REALLY bad. > At University of Bochum (W.-Germany) we use a uvax for simulations with "spice". The machine always is very overloaded with Jobs. Did anybody analyse a machine, running several spice-jobs at one time, with tools like RABBIT ? What experiences do you have ? How to change the Sysgen-Parameters ? Thanks for statements Matthias Frei (system administrator) -------------------------------------------------------------------- Snail-mail: | E-Mail address: Microelectronics Center | UUCP frei@rubmez.uucp University of Bochum | (...uunet!unido!rubmez!frei) 4630 Bochum 1, P.O.-Box 102143 | West Germany |