Re: cache-digest / b20 / wedging

--MimeMultipartBoundary
Content-Type: TEXT/PLAIN; charset=US-ASCII

> Looking at the cache log their seems to be a correlation between the
> go slow behaviour and the rebuilding of the cache digest. I was just
> wondering if this was something anyone else was seeing... Rebuilding
> without cache digest support appears to have fixed the problem.

It has not meant to be that way. Either it is a bug, or some configuration
options require tuning.

There are two constant in store_digest.c file that you might want to fiddle
with:

/* how many bytes to swap out at a time */
static const int StoreDigestSwapOutChunkSize = SM_PAGE_SIZE;
/* portion (0,1] of a hash table to be rescanned at a time */
static const double StoreDigestRebuildChunkPercent = 0.10;

Decreasing StoreDigestRebuildChunkPercent (set to 10% now) might help you.
Squid will re-calculate local cache digest using smaller chunks. Based on
your description it looks like it takes too much time to recalculate a 10%
portion of your store table. (Although I cannot believe it could take _that_
long).

I see no reason for changing StoreDigestSwapOutChunkSize, but maybe I am
missing something.

If the problem persists, please send us the cache.log entries with
debug_options for 70,71,72 modules set to level 5. This will help to pinpoint
the problem.

Thank you,

Alex.

P.S. The options I mentioned will migrate to squid.conf eventually..

--MimeMultipartBoundary--
Received on Tue Jul 29 2003 - 13:15:48 MDT