httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chuck Murcko <>
Subject Re: fairy nuff
Date Thu, 30 Nov 1995 21:34:36 GMT
Rob Hartill liltingly intones:
> > 72% vmstat
> >  procs   memory     page                         faults      cpu
> >  r b w   avm   fre  flt  re  pi  po  fr  sr s0   in   sy  cs us sy id
> >  0 0 0 21896  3780  132  15   7   0    0   0 1   258 194 110  4 11 85
> > 
> > As you can see, we're not thrashing or low on memory. The machine has 32
> > Mb RAM currently and soon will go to 48.
> This has always confused me....
> Where it says "avm", and the manpage describes this as "Active Virtual
> Memory", doesn't this mean that VM is being used and thus you don't
> have enough memory?
Yes, and this 21896k value is high for this system. normally, it runs about
an average of 11000k. defined as stuff that is running or has run in the
last 20 seconds. However, you also have to take into account the pagein (pi)
and pageout (po) rates. If they run consistently high, then real trouble is
brewing. Right now, I have 45/67 active Apache children, but I am only
paging in a little over 1 page/sec. Pageouts are currently 0, so I don't
think we're in terrible shape. I get a pageout every 2-3 minutes, hardly a
burp. Interrupts are outrunning page faults about 2:1 here currently. The
numbers above (page flt and fault in) are representative for this system. Avm
is about 10-15000k, which doesn't seem terrible for a system with this
activity. 68 Apache + 8 Netsite plus the usual minimum system daemons,
and the kernel, and the tunneler cgis (which I am now modulizing) and you
have about the right size of stuff for what I see. And, yes, we do see the
need to eventually add more RAM. Right now, this system is 32 MB Ram, 96 MB
swap. This was laid out with the thought of dropping in another 16 MB RAM
in time.

BTW, Solaris just tells you available swap, not avm, so my sense is that
the numbers mean different things on different platforms. 8^(

> When I use vmstat under SunOs (32Mb) on a busy machine avm=0
>                         HPUX  (64Mb)      quiet        avm>0
> Which confuses the hell out of me 'cos the HP machine has more memory
> and is idle when I do the test.
Sounds like you are not paging much at all on the Sun, but are on the

> So, how should one read vmstat output?
I would invite comments if my read of this stuff seems wrong. Tony?

One other weirdness is the -S flag, which changes vmstat to read swap
statws instead of page stats.

> The HP avm went down after reconfiguring the buffer cache (which is
> incorrectly configured as standard under HPUX).
That makes sense, actually, if the avm is the same as on BSDI.

Chuck Murcko	Telebase Systems, Inc.	Wayne PA
And now, on a lighter note:
This fortune intentionally not included.

View raw message