[squid-users] design or misconfiguration issue?

From: Van Bossche Koen <[email protected]>
Date: Mon, 29 Apr 2002 11:50:33 +0200

Hi all,

I have a few questions, need for advice regarding our proxy infrastructure.
I recently have done some tests on local proxies and found that they are
twice as slow as the parent caches. After I checked on the performance
issues and time synchronisation between parent and local.
I tested this on 3 local proxies, they all three are twice as slow as the
parent proxy. I believe I must have or a design or a configuration issue.

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 /...

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

When loading the second time its faster but still not as fast as from
Brussel. We changed about 250 users from 3 countries to access the parent
directly and they have excellent performance.
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 go
on having about the same performance on the local proxies as on the parent
proxies. Is this a design issue or misconfiguration?

Kind Regards,
./koen

This message has been checked by KONE Corporation for the presence of computer viruses
Received on Mon Apr 29 2002 - 05:20:52 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:07:44 MST