karl@sugar.uu.net (Karl Lehenbauer) (09/21/88)
In System V/Release 3.0 on the '386, at least, there is a configuration parameter when building kernels that says how long to hold a task out once it is swapped, to prevent thrashing. I set this to a pretty large time, two seconds, to see what it would do and because, with relatively slow 28 ms disks, the default was pretty short. Anyway, it works as advertised, but there is one, I think, misbehavior in that, the two second swapin delay occurs regardless of how long the program has been swapped out. For example, after a long period of inactivity at the terminal, if the system is doing a lot of uucp and such, my shell gets swapped out. When I come back and hit return, even if the system isn't doing anything at all, there is a two second delay before it swaps in the shell. I think the swapin holdout minimum time should be timed from when the task is swapped out, not from the time when the input request completed. Agreed? -- -- uunet!sugar!karl, Unix BBS (713) 438-5018