Re: [squid-users] Squid Performance Issues - reproduced

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Sun, 05 Jan 2003 00:25:07 +0100

Henrik Nordstrom wrote:
 
> Quick check.. cond_signal of an empty cond is a rather heavy operation
> and should be avoided if possible. Changes things a little.

Turned out to be not as heavy as first measurements indicate..

P133, Linux-2.2, glibc-2.1.3

  cond_signal 1.2us
  write(fd, "x", 1) 12us

PIII 1GHz, Linux-2.4.18-SMP (but only 1 CPU installed), glibc-2.2.4

  cond_signal 0.14us
  write 14us

but still most likely heavier than a simple counter filter within the
already used mutex..

Regards
Henrik
Received on Sat Jan 04 2003 - 16:31:27 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:19:05 MST