mark@cbfsa.att.com (Mark R Horton) (11/29/90)
Apologies in advance if this has already been covered in this group; please email responses and I'll summarize if there is interest. The Sun 4 has historically done poorly when there are many processes trying to run, because it has only 16 hardware contexts and the overhead to change processes is very high if the new process isn't one of the core 16. Thus, if the load average rises above about 18 the machine starts to have trouble. Sun says that the 470 and 490 solve this problem by increasing the number of contexts to 64 and drastically reducing the overhead to change processes. Has anyone done any experimentation on a 470 or 490 that can confirm or deny that the problem is fixed? At what point does such a machine start to have similar trouble? Thanks, Mark.Horton@ATT.COM