Unix PC disappearing clist buffers

Bill Mayhew wtm at neoucom.UUCP
Sun Apr 30 23:53:03 AEST 1989


I've been watching my 3b1, impulse, to see what it is doing.  It
seems that cfreelist+6 usually hovers around 132.  ktune -d shows
that nclist is the default of 150.

For the record, impulse has a 67 meg disk with a 5 meg swap area
and two megs of RAM.  I am running the stock 3.51 kernel.

Not having source to anything, I could only make a very wild guess
as to what happens.  I have a feeling that something must be wrong
with some sort of low-water mark algorithm that causes clist buffer
space to be reclaimed.

I modified the sysinfo program so that it shows me how much core
the kernel thinks is free in addtion to the other stuff.  After a
reboot, the free core is about 919K when I am logged in and my ksh,
sysinfo and the stuff lauched by /etc/rc are the only processes.
That amount of free core slowly disappears down to about 640K over
the course of two or three days at which time it will pop back up
to 919K give or take a couple of K.  I never tracked the clist
buffer pool before, but I intend to start watching that now too.

I am typing this article into our news machine via cu.  Before I
started cu, the clist had 132 buffers, now it has 133.  Hmmm...

For the people with source, look at the differenced between uucico
in the stock package versus HDB's uucico.  Running the stock
uucico, impulse would only stay up a couple of days before the
go-away-and-die would take place.  I haven't had any of those since
switching to HDB.  With HDB, the longest uptime I've had was 95
days, at which time I took the machine down for maintenance and
back-up.  Neoucom is our netnews 3b1; it will sporadically hang-up
every couple of weeks with the go-away-and-die.  Neouocm has a 40
meg disk, 2 meg RAM, and three tty ports, not sure about the swap
area size, also running HDB.  Impulse gets about 500K /day in mail
and news, while neoucom sees about 6-8 megs /day.  My fix for
neoucom was to build a little box with a 6502 CPU and a 2K EPROM
prgram that watches neoucom's tty001 port.  If my monitor doesn't
see the hourly poll, it reboots neoucom by opening a solid state
relay for about 30 seconds.  I did that because neoucom would
invariably go into lock-up at some convenient time like 0200!
Talke about kludge, but it is effective!  I'm thinking about adding
an airflow switch to my monitor box that cuts off the power if the
fan fails.

Bill
wtm at impulse.UUCP



More information about the Comp.sys.att mailing list