IRCG Statistics

For delivering 8GB of content (as measured by the write buffer statistics), IRCG used 16 minutes of CPU time on an Intel Pentium III at 750MHz running Linux 2.4.2. The client was connected to the server through a switched 100MBit/s network. During that time, thousands of user logins and millions of messages were handled. At peak times, more than 1900 clients were logged into the test channel and up to 20 messages were distributed to each client per second.

During a second longer run, IRCG produced 272GB of content. On Linux 2.4.2, the combined system and user time was 608 minutes which roughly translates to a net throughput of 450MB per minute of CPU time. During the test, 900 users (500 lurkers, 400 authors) were permanently connected to IRCG. More than 3.6 million messages were posted to the channel which translates to more than 3 billion delivered messages. The memory usage of IRCG peaked at 23MB.

Detailed Statistics

All runs were performed until the test client had posted at least 2500 messages to the test channel. The same test configuration as above was used. The test client was a P3-450 running Linux 2.2.19, connected through an unswitched 10Mbit/s network. We used IRCG 1.95.

Lurker Authors Total number of messages Amount of data delivered CPU time RES memory
980 20 2,435,118 454MB 4:34 19.8MB
200 50 633,440 70MB 0:36 7.5MB
300 50 865,968 101MB 0:52 9.8MB
500 50 1,433,875 184MB 2:04 13.4MB
980 50 2,811,182 440MB 2:44 20.6MB
1500 50 3,897,112 816MB 4:06 27.8MB
500 200 1,759,049 220MB 1:15 16.1MB

Notes:

Due to the way the write queue in IRCG works, it always uses a certain amount of CPU time which is not necessarily linear to the number of messages and/or receiving users. That is why the first result shows that IRCG needs more CPU time in some cases for delivering less data.

Copyright © 1999-2011 Sascha Schumann
All rights reserved.
Valid XHTML 1.1! validate | top