Re: more profiling

Here's the latest 5 minute snapshot. Squid-3 has hit the store mem limit
and is now having to delete in-memory objects to make room for incoming
requests.

(This is using datacomm-1. I wanted to see the worst-case performance -
ie, squid has to keep removing objects without getting any benefit from
HITs.)

Last 5 min averages: (Cumulated time: 1078655379460, 391.72 sec)

          Probe Name Events cumulated time best case average worst case Rate / sec % in int

PROF_UNACCOUNTED 7152744 99510227852 0 13912 424549192 18259.89 9.225
PROF_OVERHEAD 9600 1245452 0 129 88644 24.51 0.000
HttpStateData_readReply 255904 579118330532 0 2263029 23144291000 653.28 53.689
StoreEntry_write 2876390 561660160848 0 195265 23142919280 7343.00 52.070
storeGetMemSpace 2876390 522628513852 0 181695 23142884908 7343.00 48.452
comm_check_incoming 338548 328590409336 0 970587 40595044 864.26 30.463

I'm guessing readReply, write and storeGetMemSpace are nested (in that order.)
I'm curious as to why the rate/sec of readReply and write/storeGetMemSpace are
different.

I'll leave this running for 12 hours and report back.

Adrian
Received on Mon Sep 18 2006 - 21:05:57 MDT