lsf%abbott@CIT-VAX.ARPA (Sam Finn) (12/14/85)
I have run your context-switching benchmark, with the results shown below: execution times (ms): FATHER.... user: 36300 system: 40 SON.... user: 35240 system: 260 total system time: 300 i.e. 0.08 ms per context-switch Our system here at CalTech has the old (not the enhanced) CPU and 5 Mbytes of memory. As I indicated in a previous correspondence, our experience with Ridges here at CalTech seems to have been the opposite of the rest of the reporting user community. ROS is not yet a complete 4.2bsd; however, I am unaware of exceptions to 4.2, only features unimplemented at this time. It is certainly closer to 4.2 than many of the local variants and dialects that I have seen both here at CalTech and at other Universities. Our experience with multiuser performance is also better than I have seen reported by the user community. It is important to remember that a Ridge is not a Vax, that a Ridge is still a microcomputer (albeit a supermicro) with a more limited I/O bandwidth than a Vax and thus not necessarily able to service as many users as a Vax with the same level of interactive performance despite a faster CPU for number crunching. Nonetheless, a typical afternoon here will see 5-7 users on one machine (up to two coming in as remote logins over the network) working interactively. Yes, you can tell the machine is being used; no, the response time is adaquate for editing and test-running of programs. The use of our Ridges range from developement of numerical hydrodynamics and n-body codes for use on Crays, to pre- and post-processing of supercomputer data (including graphics generation), to the running of large Symbolic Manipulation Programs (i.e. smp), to a host of smaller problems whose entire life-cycle is spent on a Ridge.