jmc@root44.UUCP (07/15/83)
Has anyone ever looked particularly at 'inage' and 'outage' in sched() in slp.c in the kernel, which are used to decide which processes should be swapped out/in? These are compared with the 'magic' amount of 2 seconds, without any indication of where '2' came from. This seems to be in all USG-derived systems. Recently we had a customer (with a very slow disk) complaining about the amount of time his system took, and I put in (just out of curiousity) a hack to vary these parameters on the fly and I was STAGGERED by the difference. I got over a TENFOLD improvement in response time by upping both to 8. Any comments would be welcome. Thanks John Collins Root Computers Ltd ...!vax135!ukc!root44!jmc
lee@fortune.UUCP (07/20/83)
----- quote ----- These are compared with the 'magic' amount of 2 seconds, without any indication of where '2' came from. This seems to be in all USG-derived systems. Recently we had a customer (with a very slow disk) complaining about the amount of time his system took, and I put in (just out of curiousity) a hack to vary these parameters on the fly and I was STAGGERED by the difference. I got over a TENFOLD improvement in response time by upping both to 8. ----- I tried using 6 and 8 for inage and outage for my machine, but I didn't get much improvement over the default numbers. I guess it depends on the specify hardware configurations. Certainly, if your swapping disk is very slow, any reduction in process swapping would help. It might be better to fix the disk &/| controller &/| driver than to decrease the swapping frequency, i think. By the way, did you notice an increase in response time for more users? ( I don't have enough ports in my machine to verify this ) fortune!lee