Re: [squid-users] Questions on object access

From: Robert Collins <robert.collins@dont-contact.us>
Date: 13 Aug 2001 19:16:30 +1000

On 12 Aug 2001 22:14:28 -0500, Steve Snyder wrote:
> I noticed in my Squid (v2.4S1 with all posted patches applied) access.log
> this line which raises a few questions:
>
> 997671159.285 482 192.168.0.3 TCP_MEM_HIT/200 112615 GET
> http://www.geekazoids.net/~images/anotherpriceless.jpg - NONE/- image/jpeg9
>
> The questions:
>
> 1. Why is this a TCP_MEM_HIT given that I have this line in my config:
> "maximum_object_size_in_memory 64 KB"? Shouldn't a 112KB file be
> precluded from being found in memory given that it exceeds the specified
> 64KB limit?

Without checking the code I can't give an authoritative answer to this.
Perhaps someone else will. My *guess* is that the image had been
retrieved from the net during this instance of squid, and simply hadn't
been flushed out of memory - that is that it was still in squids hot
cache.

> 2. It took 482 milliseconds to retrieve this file from the memory cache.
> Is this reasonable for a Linux (RedHat v7.1, kernel v2.4.8) box running on
> a Pentium3/550 CPU? (With 100Mbps networking, if it matters.) Maybe my
> expectations are unrealistic. but I would think that an object retrieval
> that requires no I/O would be faster than 232KB/sec.

Weel, the 482 is elapsed time, not cpu time spent on transferring the
object. Chances are that the client couldn't accept the data faster than
232KB/sec. Also, that won't get sent in one write() call - rather a
number of calls, one per select()/poll() loop - and that will vary
widely depending on the load on your cache.

Rob

> Thanks.
>
Received on Mon Aug 13 2001 - 03:16:18 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:01:35 MST