Re: SWAPFAIL_MISS in 2.3 (nevermind)

On Fri, Jun 09, 2000, Henrik Nordstrom wrote:
> Joe Cooper wrote:
>
> > After speaking to Adrian, it looks like this is a 'feature' of the new
> > squids, wherein the cache will stop swapping to disk when overloaded.
> > Good feature...bad documentation. ;-)
>
> Hmm.. one question is how far you could push the disks before this
> feature activates in Squid-2.4. The async-io implementation most likely
> needs a bit of tuning.

Hah! I was going to post about this today! :P

I was thinking about adding a counter which each FS module could increment
whenever an open failed due to the IO levels being too high. Unless of
course, someone suggests a neat way to get feedback from store*Open for
IO failure, and to then not log swap in failures.

I'll add a section to the programmers guide outlining this. Should I
also do the FAQ too?

Adrian

-- 
Adrian Chadd			Build a man a fire, and he's warm for the
<adrian@creative.net.au>	rest of the evening. Set a man on fire and
				he's warm for the rest of his life.