2.2 vs. 2.3 service times

  I am having an interesting issue with upgrading from 2.2STABLE* to
2.3STABLE3. The compiles and everything else were flawless. I am using the
following config:

./configure \
        --prefix=/apps/squid/current \
        --enable-gnuregex \
        --enable-dlmalloc \
        --enable-async-io=64 \
        --enable-icmp \
        --enable-delay-pools \
        --enable-useragent-log \
        --enable-snmp \
        --enable-cache-digests \
        --disable-http-violations \
        --disable-internal-dns \
        --disable-ident-lookups

The issue I'm having is this: on both Solaris 7 64-bit and Solaris 2.6, my
2.3STABLE3 caches are servicing requests (both hits and misses) anywhere
between 30 and 80 ms slower than their 2.2STABLE4 counterparts (they have
identical upstream parent caches and are on the same physical subnets).
Whereas the 2.2S4 machines are taking approximately 25 requests/second and
turning them around in .18sec/.009sec for misses and hits respectively, the
2.3S3 machines are taking the same approximatel workload and turning it
around in .26sec/.03sec respectively. Has anyone else seen this type of
degradation? I have checked my sar statistics from previous days and the
rest of the environment seems to be fairly stable - if anything, 2.2S3 does
a far superior job handling file I/O (I'm using less filedescriptors and my
I/O wait percentages have decreased). I am going to try switching back and
forth a little bit to see if I can pinpoint an issue, but I thought I'd ask
the group as well.

Thanks in advance.

_______________________________________________________
Get 100% FREE Internet Access powered by Excite
Visit http://freelane.excite.com/freeisp
Received on Mon Jun 05 2000 - 16:58:38 MDT