On Tue, 8 Mar 2011 12:36:49 -0600, Baird, Josh wrote:
> Are there any docs that reference performance differences between
> 2.6/7
> and 3.1? I'm running several 2.6 clusters (forward proxy) with all
> caching disabled doing 20-30mbps per node. The nodes are not far
> from
> idle in terms of CPU and memory. They are currently running
> RHEL5/x86_64. Should I expect to see similar performance on 3.1, or
> even better?
Squid had a lot of very speed-specific changes in 2.7 that never got
ported to the 3.x series. It hit a performance high that we are still
trying to match in 3.x.
I believe 3.1 is faster than releases in the 2.6 series, but have no
solid benchmarking to back that up. The last wide comparison we had was
http://squid-web-proxy-cache.1019090.n4.nabble.com/Updated-Benchmark-results-with-CPU-usage-for-3-1-and-2-7-td1048161.html
3.1 is slower than 2.7 in most situations.
We have had one report that the very latest 3.1 releases are equivalent
under a specific type of traffic. But that was at the edge of stability
and the test machine crashed too often. It definitely takes more CPU and
memory.
3.2 surpasses everything else on the RPS metric (total parallel
requests capacity), though we know it has currently traded handling
speed in order to get there. Work is underway to gain back on the speed
regressions since 3.1, with faint dreams of beating 2.7 as well.
Its currently a matter of choice when upgrading from 2.6 and down.
If you need speed above everything else 2.7 latest is the way to go.
If you need standards compliance and new features over speed then 3.x
latest gives that
Amos
Received on Tue Mar 08 2011 - 23:07:46 MST
This archive was generated by hypermail 2.2.0 : Wed Mar 09 2011 - 12:00:01 MST