hulsebos@philmds.UUCP (Rob Hulsebos) (08/23/88)
The 'disksort()' routines I found in several kernels just sort the incoming I/O request with no preference for I/O done on swapdevices. I think it would be more efficient if any I/O to be done on swapdevices is given preference above 'normal' I/O, which can be done if disksort() checks for it, and inserts the swap I/O requests at the head of the queue. Does anybody know why this is not done ? Are there any special reasons for it ? ------------------------------------------------------------------------------ R.A. Hulsebos ...!mcvax!philmds!hulsebos Philips I&E Automation Modules phone: +31-40-785723 Building TQ-III-1, room 11 Eindhoven, The Netherlands # cc -O disclaimer.c ------------------------------------------------------------------------------
mangler@cit-vax.Caltech.Edu (Don Speck) (09/12/88)
In article <615@philmds.UUCP>, hulsebos@philmds.UUCP (Rob Hulsebos) writes: > I think it would be more efficient if any I/O to be done on swapdevices > is given preference above 'normal' I/O, which can be done if disksort() > checks for it, and inserts the swap I/O requests at the head of the queue. Why not take this further, and let I/O that some process is waiting for take priority over B_ASYNC bufs, which nothing is waiting for? This way page-ins and blocking reads would take precedence over readahead and writebehind blocks, improving response time. The CPU scheduler assigns priorities, why shouldn't the disk scheduler? Don Speck speck@vlsi.caltech.edu {amdahl,ames!elroy}!cit-vax!speck