Sample Caucus CPU Load Statistics

(Last revised 24 November 2002.)

I. Introduction
We're often asked just how much load Caucus puts on a server, or conversely, how big or fast a server one needs to buy in order to run Caucus.

The answer, of course, is "it depends" -- on how many users you expect to have, how many of them will be on at the same time, and how much activity (text and uploaded files) they will generate. 

The good news is that, in most cases, a reasonably well loaded desktop PC will do just fine as a dedicated Caucus server.  The rest of this page shows some statistics about a specific site that can be used as a guideline.

II. Summary Statistics

Site Eastern Michigan University
Sample Period October 2002
# Students 23,000
# using Caucus in the last year 11,000
# using Caucus in the last 30 days 4,165
# Active conferences > 600
Peak # users/hour 150
Peak CPU Load 20%
Disk space used 5.5 gig
Server Intel Pentium III, 933 MHz, 512 MB RAM
  Red Hat Linux 7.2, Apache 1.3

III. Graphs and Interpretation

Graph 1: Users
The first graph is the number of Caucus users on at a given time (measured once every hour).  The daily peak value is about 150, the average during working hours (which for students is more typically noon to midnight!) is closer to 75.

Graph 2: Page Views
The second graph counts the total number of Caucus pages served up to those users, per day.  Since each page is dynamically generated by Caucus CGI processes, this serves as a rough guide to the load on the server.  Notice the repeating weekly patterns.

Graph 3: CPU load
Here's the most important part of the story.  Even at peak loads of over 200 simultaneous users, CPU load never reaches 25% of full capacity.

Caveat:  it would also be useful to know how much network bandwidth was used during the peak times shown above.  Clearly, if the server was limited by network I/O, then the percentage of CPU loading doesn't mean much.  Evidence suggests that the network and network card on the server were operating well below capacity, but hard numbers are not available. 

In general, it's extremely rare for any relatively new server to be limited by network I/O.  The bottleneck, if any, is usually the CPU speed or amount of available memory.