Re: Odd swap behavior in 2.1.60

Bill Hawes (whawes@star.net)
Sun, 02 Nov 1997 13:44:10 -0500

tenthumbs@CYBERNEX.NET wrote:
> In run 2-2, I forgot to add the second swap partition so I aborted the part of
> the test that would have swapped into the second partition but I did the rest.
> I then added the second swap partition and then removed the first. As you can
> see, things happened. At system halt, new messages appeared which I added at
> the end of the file. It appears I was wrong about having to fill up the first
> partition.
>
> Run 2-3 is an exact reproduction of the original 2.1.60 problem. Again, there
> were messages at system halt which are at the end. More dire warnings, though.
>
> Now I realized I had made the same mistake as with 2.1.60 and was running a
> SMP kernel on a single-processor machine, so I built a non-SMP kernel. Run 4-1
> is the same test as 2-3. Similar results.
>
> I have run diagnostics on the RAM and the drive and there are no errors so
> this looks like a real problem.

Thanks for the error logs -- it does look as though your system has lost
track of swap pages, lots of 'em. More than 10M ...

> Let me know what you want to try next.

OK, I'd like to see if there's some pattern of usage associated with
it. Try turning off the first swapfile soon after it first starts
swapping, then turn it back on. Keep trying that until you start
getting the page unused messages, and see if you can find an unusual
event that causes the pages to be lost.

Regards,
Bill