hit rates and maximum_object_size

From: Neale Banks <neale@dont-contact.us>
Date: Sun, 12 Dec 1999 08:57:45 +1100 (EST)

Hi,

Having set up a new squid (Debian package 2.2.4-1.slink (i.e. the potato
package recompiled for Slink)), the hit rates were somewhat lacking (e.g.
~15% by bytes), and there is pressure from above to get it up to something
repectable (e.g. closer to the "mythical 30%" discussed here last month).

So the maximum_object_size (previously default, 4096 KB) was a few days
ago explicitly set:

        maximum_object_size 32768 KB

and a "squid -k reconfigure" issued.

The theory was that this move would _increase_ the hits-by-bytes (based on
the comment in squid.conf: "If you wish to get a high BYTES hit ratio, you
should probably increase this").

It's now several days later: Calamaris (2.27) has been reporting a
steadily declining hits-by-bytes and is now down to 10% :-(

Snippet from the last 24hours:

----------------------------------8<----------------------------------
# Incoming TCP-requests by status
status request % kByte % sec kB/sec
--------------------------------- -------- ------ --------- ------ ---- -------
HIT 107201 30.24 281595 10.08 1 3.31
 TCP_HIT 29564 8.34 170219 6.09 1 5.40
 TCP_REFRESH_HIT 26192 7.39 85043 3.04 2 1.65
 TCP_MEM_HIT 6086 1.72 12751 0.46 0 257.42
 TCP_IMS_HIT 43600 12.30 12391 0.44 0 6.52
 TCP_NEGATIVE_HIT 1759 0.50 1192 0.04 0 83.92
----------------------------------8<----------------------------------

Am I doing something wrong here?

Thanks,
Neale.
Received on Sat Dec 11 1999 - 14:45:09 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:49:50 MST