ES Designs Logo

Webserver Comparison

Updated: 3/2/97


I started playing around with the MS Personal Web Server last weekend. I was particularly interested in seeing how it compares to Quid Pro Quo. I have been using QPQ for several months now for testing and demoing CGIs.

With the release of MSPWS 1.0, another free webserver. I figured that it was a good time to do a comparision. First some general commments about MSPWS. Setting up and running the server was very easy. Since MSPWS uses the Mac's file sharing preferences, you don't have to worry about setting up realms or passwords. (This appears to be a two-edged sword, since you can apparently only set up a single realm for the drive owner).

Testing

I wanted to some measure of server performance, an idea of how long it takes the server to respond to an request and deliver html. To do this, I timed the response time to a Frontier CGI. I started a stopwatch when I clicked on a submit button and stopped it after the subsequent page was delivered. The resulting page is small 2.4K so its load time should be relatively small. I made 20 runs for each test to diminish errors caused by varying reaction time and transient system variations.

All the tests were run on a PM 8500/120 80Megs RAM, Sys 7.5.5, OT1.1.2, and OT/PPP 1.0f1c9. The server and browser were running on the same machine. The results that I got appear to be internally consistent, but given the obvious limits of this test, they should not be taken as gospel, YMMV.

The results were interesting. I initially tested Quid Pro Quo (QPQ) and MS Internet Explorer (IE). I then tested MS Personal Web Server (PWS) and IE and the responce time was consistently faster than QPQ/IE averaging 4.22 seconds compared to 5.37 seconds.

During the testing, I noticed that QPQ closed connections much faster than PWS. The connection in QPQ closed almost imediately after the page was served, but remained open for several seconds after the page was served in PWS. I wondered if this might have an effect on the server speed, so I retested PWS but this time waited until all connections were closed before sending the next request. This new test revealed a potential bug with PWS. With no open connections, the first request failed. The subsequent request went through with a response time similar to those observed in the PWS/IE test described above (average response time 4.15 seconds over four sucessful runs).

I then started wondering whether using Netscape Navigator 3.0 (NN) as the web browser would make any difference, so I repeated the tests with QPQ/NN and PWS/NN. These results were similar to those observe with Internet Explorer (IE) with PWS serving the file over a second faster than QPQ (the response time for PWS/NN averaged 2.71 seconds compared to a 4.03 second response time for QPQ/NN). Waiting until all the connections were closed in PWS did not adversely affect the response time (2.72 seconds), more significantly there were no failures This indicates that the bug observed with in the PWS/IE test is related to this combination of products.

The following table shows the results of these tests.

RunQPQ/IEQPQ/NNMSPWS/IEMSPWS/IE (wait till 0)MSPWS/NNMSPWS/NN (wait till 0)
1 5.99 3.17 3.84 failed 2.46 2.56
2 6.03 3.9 4.24 3.97 2.4 2.53
3 5.21 4.28 4.09 failed 2.67 2.74
4 6.59 3.46 3.81 4.37 2.84 3.06
5 4.67 3.75 3.8 failed 2.46 2.71
6 5.15 5.22 4.28 4.75 2.4
7 5.03 4.1 4.53 failed 2.75
8 5.05 6.78 4.46 3.5 2.71
9 5.55 3.21 4.34 failed 3.1
10 5.4 7.28 3.84 mis-timed 2.5
11 5.12 3.46 3.65 failed 2.28
12 5.28 3.46 4.34 3.15
13 6.78 2.96 4.54 2.59
14 4.72 4.46 4.09 3.09
15 5.4 3.4 4.28 2.73
16 6.05 4.21 3.91 2.84
17 5.23 3.23 4.37 2.71
18 4.34 3.56 4.03 2.8
19 4.59 3.46 4 3.21
20 5.16 3.34 5.92 2.43
Average: 5.37 4.03 4.22 4.15 2.71 2.72

Discussion

Without further testing on a number of systems, it would be difficult to figure out the real significance of these tests. Data collected here may not even be applicable to a real world setting, where data are served over the Internet. They do, however, suggest that the MS Personal Web Server is significantly faster at processing requests and delivering web pages than Quid Pro Quo on a local system. This could be important if you are developing and testing CGIs. It could also be important in setting up a web browser interface to applications.

On the down side, PWS does not appear to have sufficient security controls to be used for anything other than a personal web server. PWS appears to only provide a single realm for the entire site and a single username and password to access the realm. Since this is the same username and password that you give to control access to your hard drives, distributing it to others who may access the site is an unacceptable security risk.

For now, the ES Design's Test Server will continue to be run from Quid Pro Quo. It remains the best full-featured web server for the price.


[Home]

Waiting for that??? | System Stability | Webserver Comparison

by Tom Clifton, © 1997 ES Designs
email: tclifton@es-designs.com