Squid Performance

Hello,

we are testing squid-2.2.stable5 on multiple platforms for use as an
application
gateway (proxy) in a firewall environment.
!!! Cache functions are not used.
Our goal is to achive a maximum of performance.
The performance tests are made with WebStone 2.5 with a client range of 150
to
1000 client processes.
The test-load is like HTTP traffic :
25% Files 500 Byte
25% Files 1 kByte
15% Files 3 kByte
25% Files 5 kByte
5% Files 10 kByte
5% Files 50 kByte.

After we tried SUN Solaris machines (U60, 2x450MHz, Solaris 2.6) with
unsatisfactory
results...

SUN Ultra 60:
Clients ConnectRate ConnectTime RespTime ErrorRate Troughput
                conn/s msec sec % Mbps
150 245,3 0,44 0,61 0,135 10,8
200 243,5 0,46 0,81 0,135 10,7
300 248,1 0,43 1,20 0,189 11,0
400 242,7 0,43 1,62 0,198 10,8
500 238,8 0,45 2,05 0,191 10,6
600 211,2 1,38 3,25 0,186 9,4
800 200,6 3,41 3,91 0,204 8,9
900 189,0 5,26 4,67 0,212 8,4
1000 181,1 7,35 5,43 0,204 8,0

(SUN specialists told us that we've used all Solaris tuning possibilities)...

...the next test was made on a Dell GX1(400MHz PentiumII, Linux 2.2.13).

PentiumII:
Clients ConnectRate ConnectTime RespTime ErrorRate Troughput
                conn/s msec sec % Mbps
150 334,1 0,91 0,44 0,096 14,8
200 325,5 1,02 0,61 0,073 14,4
300 311,4 0,88 0,96 0,057 13,8
400 296,1 4,33 1,35 0,052 13,2
500 281,0 13,61 1,78 0,041 12,4
600 268,2 18,52 2,24 0,033 11,8
700 257,2 27,06 2,72 0,034 11,4
800 248,0 42,20 3,24 0,027 11,0
900 240,2 55,74 3,77 0,023 10,6
1000 233,5 89,59 4,31 0,023 10,3

And then we tried a AMD K7 700MHz on a Gigabyte Motherboard. A Compilerrun
of a
2.2.13 kernel took half the time then the run on the Dell.

K7:
Clients ConnectRate ConnectTime RespTime ErrorRate Troughput
                conn/s msec sec % Mbps
150 452,4 1,51 0,33 0,103 20,0
200 436,9 1,55 0,45 0,094 19,4
300 407,3 1,23 0,73 0,082 18,0
400 380,4 3,97 1,05 0,065 16,9
500 354,8 10,31 1,40 0,058 15,7
600 333,7 15,11 1,79 0,047 14,8
700 314,0 19,80 2,23 0,041 13,9
800 296,9 34,01 2,70 0,036 13,1
900 282,9 46,86 3,19 0,032 12,5
1000 271,0 65,75 3,71 0,033 12,0

But the test of squid on th k7 machine showed a decrease from 20 Mbps at
150 Clients
(150% of P2 peformance) to 12 Mbps at 1000 Clients (only 116% of P2
performance).
I think it should be better.

Whats your meaning about that? Are there some hints specially related to
the K7 ?
Are there other bottlenecks ( I think it cannot be disk: no cache, logs
>/dev/null) ?

Frank Tiede

#squid.conf - a very basic config file for squid

http_port 80

cache_access_log /dev/null
cache_log /var/log/squid/cache.log
cache_store_log /dev/null

#Turn logging to it's lowest level
debug_options ALL,1

#defines a group (or Access Control List) that includes all IP
#addresses
acl all src 0.0.0.0/0.0.0.0

#acl XXX src 192.168.3.0/24
no_cache deny all

#allow all sites to use connect to us via HTTP
http_access allow all

#allow all sites to use us as a sibling
icp_access allow all

#test the following sites to check that we are connected
#dns_testnames internic.net usc.edu cs.colorado.edu mit.edu yale.edu

#run as the squid user
#cache_effective_user squid squid
#otherwise, you can uncomment the below line and comment the one out.
#
cache_effective_user squid nogroup
Received on Tue Jan 11 2000 - 08:12:10 MST