Return to BSD News archive
Newsgroups: comp.unix.bsd Path: sserve!manuel!munnari.oz.au!network.ucsd.edu!sdd.hp.com!elroy.jpl.nasa.gov!ucla-cs!ucla-mic!pita!scott From: scott@pita.cns.ucla.edu (Scott Burris) Subject: Re: mysterious system hangups (W/FIXES!) Message-ID: <scott.713466733@pita> Nntp-Posting-Host: pita.cns.ucla.edu References: <1992Aug4.175738.7008@Unibase.SK.CA> <michaelv.713151876@test.cc.iastate.edu> <1992Aug8.073502.13655@Unibase.SK.CA> Date: 10 Aug 92 10:18:36 PDT Lines: 54 roe@Unibase.SK.CA (Roe Peterson) writes: >michaelv@iastate.edu (Michael L. VanLoon) writes: >>In <1992Aug4.175738.7008@Unibase.SK.CA> roe@Unibase.SK.CA (Roe Peterson) writes: >[ summary of my summary deleted :-) ] >>I'd just like to make it clear that not all machines suffer from this. >[...] >>I've pushed the machines VERY hard and have had NO problems whatsoever. >>I am happy to report total and complete stability. >(Aside: how much memory in the machine?) >Well, after having applied two fixes, I'm _also_ happy to report total >and complete stability. Seems one fix is a work-around that is >the subject of 386BSD 0.2, and the other just makes sense. >If you've a lot of machine memory, (ie: > 8MB), do this: >> From James da Silva: >> >> "I don't think this is related to the NFS patch. Bill posted >> about this problem last week. A kernel table is sized too small for >> active machines with large (12MB or more) memory. Try modifying the >> value of MAX_KMAPENT in /sys/vm/vm_map.h from 500 to 1000. It worked >> for me." >ALWAYS do this (courtesy of W. Jolitz himself): >> Insert before line 170 of machdep.c the following line: >> >> bufpages = min(NKMEMCLUSTERS/2, bufpages); >(I believe the file is /sys/i386/i386/machdep.c) >These two changes have produced a completely robust system. I've rebuilt >the kernel, libc.a, and a whole whack of my own code simultaneously, with >no problems whatsoever. >Nice job, Bill. Unfortunately, these two patches don't make a robust system in all cases. I've applied both patches to a 16MB 486 system and although it has improved things considerably, I've still hung in the memory allocator and I've seen one kmem_map too small panic. These problems occur under heavy SCSI activity to a disk and a CD-ROM drive. -- ---------- Scott Burris UCLA Campus Network Services cnetslb@oac.ucla.edu (213) 206-4860 - OR - scott@pita.cns.ucla.edu