Re: Slowness of Squid without any apparent reason

On Friday 18 October 96, at 15 h 29, the keyboard of jrg@demon.net (James
R Grinter) wrote:

> But the size of the process is odd for the amount of accesses you
> have. Are people transferring really large objects (use cache
> manager and look at the VM objects list and total up the sizes)?

There are 180 such "in transit" objects, including three different
Netscape Navigator 3 for MS-Windows (with different URLs, so Squid
doesn't notice it's actually the same) and two for Sparc. Many PDF,
PostScript files, etc, too.

url_get cache_object://localhost/stats/vm_objects | awk '{sum += $3} END
{print "Sum: ", sum }'

tells us that we have 80 megabytes in VM objects. It can explain at least
a part of the problem.

To summary:

cache_mem (objects "cached" in memory) = 130 Mb
VM objects (being retrieved) = 80 Mb
Squid data structure for objects in the disk cache : 50 Mb (according to
Squid in "StoreEntry", we have 800 000 objects on disk)

Since my Squid is 310 Mb, I still miss 50 Mb :-} The various data
structures used by Squid and displayed at the end of the "info" output
seem to be large enough to explain it. If I understand well, our only
choice is to drastically reduce cache_mem or to buy more memory?
Received on Fri Oct 18 1996 - 08:05:59 MDT