🚀 go-pugleaf

In article <3342@blake.acs.washington.edu> lgy@blake.acs.washington.edu (Laurence Yaffe) writes:
>...This response ("buy more memory, or sequence jobs by hand"), while common,
>completely misses the point.  The real issue here is how to maximize overall
>performance with a given physical configuration...

Well, but I don't think it *entirely* misses the point.  One should be
aware that there are circumstances in which *no* amount of fiddling will
get reasonable performance out of a given configuration and a given
workload, as the two are simply incompatible.

>    As to your specific question - yes, I have considered how long it takes
>to swap out a 25 Mb process.  On my system, perhaps 10-20 seconds.  This is
>utterly negligible for jobs which require many cpu-hours or cpu-days.  

It would have been nice if you'd specified that these are background compute
jobs, not interactive applications; your example (Mathematica) strongly
suggested otherwise.  Timesharing machines and batch computing servers
are very different applications of the hardware.  For interactive processes,
I stand by my original comment:  the system is simply too small.  For batch
work, the idea is reasonably practical but requires very different system
tuning, which Unix (traditionally an interactive system) hasn't given a lot
of attention to.

Also, have you *timed* that swapping, or is that just maximum disk transfer
rate?  Remember that your system probably can't drive your disks at full
speed; disk transfer rates are "guaranteed not to exceed" rates, not often
achievable in practice.  Your point is still valid, in that the time is
negligible for long-running batch jobs, but a dedicated batch machine is
clearly needed, as that kind of disk activity will ruin interactive use.
-- 
V7 /bin/mail source: 554 lines.|     Henry Spencer at U of Toronto Zoology
1989 X.400 specs: 2200+ pages. | uunet!attcan!utzoo!henry henry@zoo.toronto.edu