Subject: Re: BUFFERCACHE, PR 1903
To: None <current-users@NetBSD.ORG>
From: Laine Stump <laine@MorningStar.Com>
List: current-users
Date: 09/14/1996 20:48:31
Okay, all this talk about parameters I've never heard of ha smy interest
peaked. Who has a good recommendation for settings of kernel options for
a machine with 128MB of RAM that is used for 1) managing the X desktops
of 4 engineers (including running emacses, etc), and 2) doing lots of
BIG compiles (800,000 lines * about 20 platforms) which are run with
make -j<large number>?

The reason I'm curious is that we are once again having problems with
occasional 15-20 second "lockups" of the machines when they start
swapping. (the amount of free mem gets lower and lower, then all
processes stop, waiting on "thrd_s" while the disks get accessed about
once or twice a second for awhile (and nothing else happens), then
finally everything starts back up for awhile). I'm thinking that maybe
different settings for cache'y memory'y things might make it start
swapping more sooner, avoiding (or reducing) the time it spends in this
degenerative state. (Actually, having it wait until so late to swap
wouldn't be a bad thing, if it just wouldn't sit doing *nothing* for a
second between each disk access - I'd sure like to know what it's doing

The last time I solved this problem, I did it by upgrading our two
machines from 64MB to 128MB. Now we're uing them for more stuff, so
128MB isn't enough either and I'm about to make them 256MB.

These are both P6-200s with 2940UW SCSI controllers, dual 4.3GB FW disks
(striped) (raw disk throughput is about 11Mbytes/sec) connected to
100Mbit ethernet. They really scream until they swap, then they suck
eggs. Last time I posted about this, someone said they were going to
start working on it; any progress there? This is a real annoying problem
(wish I had the knowledge/time to fix it). Soon people in the office
will start listening to the office Linux fanatic.