Subject: Re: memory tester shows up swap/page tuning bug ...
To: None <Havard.Eidnes@runit.sintef.no, jmarin@pyy.jmp.fi,>
From: Sean Doran <smd@cesium.clock.org>
List: current-users
Date: 09/14/1996 19:19:02
>What kind of paging is being seen during all of this?

A steady, constant, 64 pages/sec. That's what the 64 disk xfers/sec
was, sorry for not pointing that out.

        PAGING   SWAPPING      Interrupts
        in  out   in  out           total
count        65    
pages        64

I sent a PR on this, perhaps someone who knows the NetBSD VM system
(e.g., perhaps Charles) rather than the 4.3 vm system (Sean and I)
will take a look at this.

It looks like nothing much at all is happening during the freeze
periods, the freepage list is exhausted, pages are being dribbling out
to disk at a low, constant rate --suspiciously close to "scan", which
is 65 -- until _something_ magic happens, the freelist goes nonzero,
and the page-toucher goes back to happily thrashing at a rate of
1.5Mbyte/sec of disk I/O.  The disk on this machine should sustain
that forever, if it wasn't for the freezes.

The 1.5Mbytes/sec is very close to twice the rate at which the
memory-toucher makes progress when it's not frozen. I conclude the
memory-toucher is touching ~0.75 mbytes/sec, reading in that many
pages and also writing out that many pages to make room for the ones
paged in.  (It runs much faster  at startup when essentially  all
of memory is free.)

Now that there's a simple way to produce this behaviour on demand, and
at least one tool that shows what's going on, it should be
straightforward for someone who knows the VM system well to find the
cause of this and fix it.

--Jonathan