Forum OpenACS Q&A: "Oh my goodness! AOLserver 3.0 is one seriously fast beast."

Here is excerpt of benchmarks comparing aolserver 3.0 to apache 1.3.12 and Zope 2.2b2:
Oh my goodness! AOLserver 3.0 is one seriously fast beast.

   Here are benchmark results returned by ab, the benchmark utility
that is part of the Apache distribution, running on RedHat 6.2, on a
Dell P3-700 with
   256 MB of RAM.

   Zope 2.2b2:
   50 concurrent users
   700 requests
   5.85 seconds to process
   119.66 requests per second
   172.07 kb/s transfer rate

   Apache 1.3.12:
   50 concurrent users
   700 requests
   1.805 seconds to process
   387.81 requests per second
   731.95 kb/s transfer rate

   AOLserver 3.0:
   50 concurrent users
   700 requests
   0.525 seconds to process
   1,333.33 requests per second
   2,383.60 kb/s transfer rate

   Of course there is more to a Web server than just how fast it is.
But speed is certainly one very important part of a Web server.

       
Oh I forgot.  The link is at http://weblogs.userland.com/qube/2000/06/26, and it is linked from http://www.apacheweek.com.  It sounds like it was for static page returns, so it's probably not so meaningful.  It would be nice to see a similar test which involved database accesses.
Stay tuned: the guy who did the benchmark says in a later entry that he wasn't expecting all the attention that he got for his benchmark and so he expects to do a more serious one later on that also checks dynamic pages, etc.
Collapse
4: More benchmarks: (response to 1)
Posted by Jonathan Ellis on
I ran some more benchmarks, also using ab, since I wanted to see a comparison against IIS. Since the original author didn't specify what page he was serving, I ran mine against the default "It Worked!!" apache page from RH 6.2. (Which has an extra .png image from the regular apache "It Worked!!" page, FWIW.)

I also got wildly varying results testing with 700 requests so I tested with 10000 instead. I kept concurrency level at 50. These results are for PIII/450 single processor servers with 256 MB of RAM.

Short version: IIS kicked AOLServer's and Apache's butts. (Boo hiss!)

On Apache:

Requests per second:    718.39
Transfer rate:          1587.64 kb/s received
Connnection Times (ms)
              min   avg   max
Connect:        0     0     5
Processing:     8    68   618
Total:          8    68   623

On IIS:

Requests per second:    1450.12
Transfer rate:          3143.85 kb/s received
Connnection Times (ms)
              min   avg   max
Connect:        0     1  3280
Processing:     9    32    35
Total:          9    33  3315

On AOLServer:

Requests per second:    745.66
Transfer rate:          1605.40 kb/s received
Connnection Times (ms)
              min   avg   max
Connect:        0     0     5
Processing:    14    52  3055
Total:         14    52  3060

I'm trying to email the author of the linked article asking what page he was serving. I'm curious how he got results of AOLServer being so much faster than Apache. (I'm as willing as the next guy to believe the AOLServer's fast, but I'm just not seeing the kind of difference he was claiming.)

I'm also planning to do a test of dynamic pages with IIS + ColdFusion + Oracle vs AOLServer + Oracle. (Any suggestions of what you'd like to see tested? Please include tcl. :)

BTW, many thanks to the authors of the new OpenACS install guide. BIG help!

Did you do these benchmarks using NT or Win2K or what ?

You can expect a benchmark from me in the next few weeks using Traffic Jamme, that should do some real testing. BTW, I think benchmarks are more realistic for big websites if done in SMP boxes.

On the documentation, you're welcome. I am glad people are liking it. You can help us keep it up-to-date by submitting bugs and feature requests at the SDM.

The IIS numbers are from a machine running NT 4.0 SP6.  The others are from a machine running RH 6.2.

I ran these on our single processor machines so the numbers would be roughly comparable to the Qube guy's hardware.  Shrug.

I am kind of suspicious about benchmarks running on different machines... Also, you seem to be a NT-type of person (as shown by you having the latest SP from Microsoft), so you probably did not tune your Red Hat install as carefully (if any) as your NT box.

For example, I can compile a Linux kernel to be much more optimized than the default that comes with Red Hat. Any serious Linux user knows that.

I doubt you'll get better results with ColdFusion, since AFAIK, it's process-based and forking a backend for each request.

Note that these are strictly technical considerations.

Man, I never thought I'd be mistaken for an NT nerd. :-b

I only ran them on different machines because IIS doesn't run on Linux (duh) and last I heard Apache on NT underperformed Apache on Linux.  Shrug.  The hardware is, like I said, identical.

(Yes, the RH box is running a custom kernel.)

As a follow-up to this, the Qube guy re-checked his benchmarks, figured out he'd screwed up badly, and came up with new results showing in essence a dead heat between Apache and AOLserver in the serving of static pages.
Seeing his new test run, what is the matter with AOLserver 3.0 and keepalives?
Stop by the chat with the AOLserver developers today and ask them. I will. Instructions are at community.aolserver.com.