Most Current Squid Benchmarks

Speed and Requirement details of squid are a little hard to come by at present. Here is a list of the community contributed achievements.

If you are running any release of squid and can provide the same details with a better requests-per-second than one listed we would like to know about it.

Sorted by Squid Release and CPU.

Method of Calculation

There is no good fixed benchmark test yet to measure by so comparisons are not strictly correct. Here is how the follow details are calculated:

• At an administrator estimated peak traffic time run squidclient mgr:info or otherwise pull the info report from cachemgr.cgi

Records

Squid trunk revno 13251

This number was taken in a controlled test environment. It has nothing to do with the numbers someone would get in a production environment; it's just an estimate of how fast squid can be. Squid was configured to do no logging, no access control, and apachebench was used to hammer squid asking 10M times for a static, cacheable, 600-bytes long document. Of the 4 cores, 3 were running a multi-worker squid, one was running ab over the loopback interface.

Submitted by: FrancescoChemolli 2014-01-30

Squid 3.2

3.2.0.14

Submitted by: Squid BuildFarm 2012-01-18

This is achieved using laboratory tests with Web Polygraph for raw request throughput with no SMP workers engaged.

This is published as a theoretical median for high performance. Faster/newer hardware will reach higher throughput at peak and most networks will not reach this rate in real production traffic.

3.2.0.9

Submitted by: Jenny Lee 2011-07-03

Client database is disabled. Proxy-only, no caching done. 5 to 20 parents. Squid doing content routing. conntrack_max and hashsize increased to 200K, ephemeral ports 64K. tcp_tw_recycle on.

Squid 3.1

3.1.10

Submitted by: Jack Quinlin 2010-12-29

Squid 3.0

STABLE 5

Submitted by: Philipp Rusch - New Vision. 2008-07-17.

IBM xSeries 3250 M2. This system is doing virus-scanning with ICAP-enabled Squid through KAV 5.5 Kaspersky AntiVirus for Internet Gateways
AND it is doing web-content filtering with SquidGuard 1.3
AND it is doing NTLM AUTH against the internal W2k3-ADS-domain

Unknown Release

Submitted by: Krzysztof Dajka. 2010-05-24

Running squid3.0 on Dell R300 servers. On production servers getting max 1500hits/s. With 2500hits/s I have seen that some in access.log, in elapsed column that some requests were closed after 6 seconds and average was something like ~300ms.

Squid 2.7

unspecified

Submitted by: Gareth Coffey. 2012-03-20

STABLE 9

Submitted by: Quin Guin. 2010-12-02.

We handle mostly mobile HTTP traffic so small files.

STABLE 7

Full Details: http://www.squid-cache.org/mail-archive/squid-users/201002/0838.html

Submitted by: Markus Meyer. 2010-02-25.

We have to handle mostly very, very small files which is
a real pain. So COSS was my white knight to handle this.

Although we don't use CARP we made sure that the proxies always get the
same requests. We have at our prime time about 40 MBit/s outgoing
traffic which makes about 1000 requests per second.

Also I should mention that we use a standard Debian kernel with no
tuning in any kernel parameters except the following two:
  net.ipv4.tcp_max_syn_backlog = 4096
  vm.swappiness = 0

STABLE 6

Submitted by: Quin Guin. 2009-04-07.

We handle mostly mobile HTTP traffic so small files.

CPU Usage, 5 minute avg:    4.33%
CPU Usage, 60 minute avg:    3.97%

STABLE 4

Submitted by: Nyamul Hassan. 2008-11-18.
Squid is doing a close to default configuration with ICP with peers and Collapsed Forwarding off.

Squid 2.6

STABLE 21

Submitted by: Quin Guin. 2009-04-07.

We handle mostly mobile HTTP traffic so small files.

CPU Usage, 5 minute avg:    10.45%
CPU Usage, 60 minute avg:    10.21%

Submitted by: Jan-Frode Myklebust. 2009-04-06.

STABLE 18

Submitted by: Gavin McCullagh, Griffith College Dublin

Cache: 1x 600GB. With this hit ratio and cache size, substantial cpu time is spent in iowait
as the disk is overloaded.  Reducing the cache to 450GB relieves this, but
the hit rate drops to more like 10-11%.

STABLE 6

Squid 2.5

NP: probably 2.5.STABLE7 or earlier going by the release dates.

Submitted by: Martin Marji Cermak. 2005-01-14.
http://www.squid-cache.org/mail-archive/squid-users/200501/0374.html

Other Benchmarking

SquidBlocker 2015

The test is for one proxy which runs a url filtering DB helper. The helper runs http queries against one reverse proxy which RoundRobin load balance to three backend DB servers. The squid settings:

workers 2
visible_hostname proxy2

external_acl_type filter_url ipv4 children-max=20 children-startup=10 children-idle=5 concurrency=50 ttl=3 %URI %METHOD %un /usr/bin/sblocker_client -http=http://filterdb:8082/sb/01
acl filter_url_acl external filter_url
deny_info http://ngtech.co.il/block_page/?url=%u&domain=%H filter_url_acl

acl localnet src 192.168.0.0/16

http_access deny !filter_url_acl
http_access allow localnet filter_url_acl
access_log none

Environent topology at: http://ngtech.co.il/squidblocker/topology1.png

This number was taken in a controlled test environment. It has nothing to do with the numbers someone would get in a production environment; it's just an estimate of how fast squid can be. Squid was configured to do no logging and apachebench was used to hammer squid asking 250K times for a blocked url (leading to a 403 response with a location header) or with a cacheable, 16KB long document. Of the 4 cores, 2 were running a multi-worker squid. The apache benchmark was run from another host and from the same host with similar results.

Submitted by: Eliezer Croitoru 2015-08-25

Older

Mark Nottingham benchmarked Squid 2.5 vs 2.6 in late 2006: http://www.mnot.net/blog/2006/08/21/caching_performance

The Measurement Factory benchmarked Squid 2.4, in particular IO systems in 2000 http://polygraph.ircache.net/Results/bakeoff-2/

Bryan Migliorisi posted some analysis of Squid-2.6 speeds under pressure on September 8th, 2009 at unfortunately a dynamic website that disappeared on us.

---

CategoryKnowledgeBase