Re: SWAPFAIL_MISS in 2.3 (more info)

Joe Cooper wrote:
>
> This has been discussed a couple of times by other folks in the past two
> months but no solution has been presented. So I thought I'd bring it up
> again to see if anything has been figured out.
>
> In 2.3.STABLE3 (and STABLE2 as well...not sure about STABLE1), an
> async-io compile will begin losing it's cache and reporting
> SWAPFAIL_MISS entries after a short time under load. If I run for a few
> hours under load, the hit rate will drop from (for example, on a
> Polymix-2 workload) 55% to 30% (mostly MEM_HITs).

In order to, hopefully, help with debugging, I've gathered a little more
information.

I upped the debug level to 10 and then grepped out everything related to
one of the SWAPFAIL entries' URI just to see if I could figure out
what's happening, and I've posted the log entries for each of the three
logs on a webpage (kinda big to post here)...

http://www.swelltech.com/pengies/joe/swapfail.html

I hope this makes the bug easier to chase. If I can provide any more
information, please let me know.

Thanks for any input folks can provide.
                                 --
                    Joe Cooper <joe@swelltech.com>
                Affordable Web Caching Proxy Appliances
                       http://www.swelltech.com
Received on Wed Jun 07 2000 - 22:51:56 MDT