[squid-users] configuration or design issue?

Hi all,

I posted this message a few days ago. I would appreciate getting some advice
about our configuration.

I recently have done some tests on local proxies and found that they are
twice as slow as the parent caches.
I tested this on 3 local proxies, they all three are twice as slow as the
parent proxy.

I am not really sure where the problem might be located :
Some explanation about our infrastructure :
We had to implement proxies because we cannot have a default route to the
Internet (using Equants network).
We use for Europe about 10 local proxies (country based) and 2 parent caches
in Brussels.

The parent caches are configured with 10Gb cache, SCSI (mirror), Diskd,
Squid2.5DEV, RedHat v7.1
parent squid (configured with redirector squidguard) -> trend interscan ->
FW -> Internet access point Equant

The local proxies use a local cache of 4Gb (mostly IDE based), diskd,
Squid2.5PRE5, Redhat v7.1
They are configured in a round-robin way to access one of the 2 parent
caches using ICP :
cache_peer IP_parentproxy1 parent 8080 8081 no-digest no-netdb-exchange
round-robin
cache_peer IP_parentproxy2 parent 8080 8081 no-digest no-netdb-exchange
round-robin

I have regarding performance installed bind+local_caching_ns / tuned
smb_auth / checked with aufs in stead of diskd /... Installed the
timeservers for timesynchro between local and parent / checked their local
network / ....

Some measurements :
loading www.tnt.com on parent proxy takes about 14 seconds
loading www.aftonbladet.se on parent proxy takes about 20 seconds
loading www.cnn.com on parent proxy takes about 36 seconds
loading www.tnt.com on local proxy takes about 38 seconds
loading www.aftonbladet.se local proxy takes about 43 seconds
loading www.cnn.com on local proxy takes about 1 minute 10 seconds

We changed about 250 users from 3 countries to access the parent directly
and they have excellent performance using it this way.
MRTG runs on those boxes and tell me their cache hits are 45%. The local
proxies have about 140, 75 and 50 users.

Please from your experience could you give me some hints, advice how to
resolve this performance issue. Latency is ok (they have no problems using
the parent1), is it because not to many users use those local proxies. Half
the performance is really a big difference, because of this also their SSL
connections break.

Kind Regards,
./koen

This message has been checked by KONE Corporation for the presence of computer viruses
Received on Thu May 02 2002 - 08:33:53 MDT