---- snippet from Steve Willis responding to my email --------
>> Performance of a robust web server ITK is quite appreciable. I would
>> suggest taking a look at the TCP Server Deux demo with A4D and run it
>> compiled. You will see then some real performance...
>
>I did and it was faster than ITK but still slower than the 4D Web Server.
>
>My testing is by no means scientific but the differences are obvious. I took
>Aparajita's demos, added a bunch of images to queries.a4d; did the tweak to
>the ITK code Randy suggested; and turned the web server cache off in 4D web
>server. I then ran the 3 demos (compiled) in the same environment and found
>the following approximate response times.
>
>ITK - 5-6 seconds
>TCP Server Deux - 3-4 seconds
>4D web server - 1-2 seconds
>
>This surprises me because past experience with ITK has always found it very
>fast.
I wanted to go and get some numbers on this here before speculating
anymore on this subject. We obviously run a lot of servers here, most of
which we developed ourselves and which use our components (all using ITK).
We added some new features to TCP Server Deux for the next beta release. Basically, the new features allow for timing logs to be generated directly by the server. The locations for pulling timing stats. are at critical points in the stream handling processes within the server. Of course, the timing logs themselves slow down the server appreciatively. But, it still gives a good idea of where slowness might be within different systems.
We found that the actual times are still nothing like what is mentioned above. We ran some of these tests to even push the server _over_ what should be done...
Let me explain... We have posted three test results for you to look at:
a) TCPsd_TL_browser.txt: this is a baseline test of just a single
browser window repeatedly hitting a server. This is a good basis of
numbers to work from for reference and it shows how much time logging
slows down the TCP Server Deux component (time logging adds approximately
80ms to each total connection time). More details of the test are on the
result page (see the link, below).
b) TCPsd_TL_splat_bad.txt: this is a test to overload the server right from the beginning and to keep it overloaded for a period of time. Basically, the number of listeners and handlers is kept constant. The number of connections is double the amount of listeners and handlers, a situation which should be rectified by increasing the number of listeners/handlers in the server. Basically, this shows what over-utilization of poor server settings will do and what kind of response can still be expected from the server.
c) TCPsd_TL_splat_edge.txt: this is a test to push the server right to
the edge and see what happens. The number of listeners and handlers is
kept constant, the same as the number of simultaneous connections being
repeatedly made. This shows what the server would do when it is "right at
the edge" of acceptable performance for utilization load it is receiving.
JIC, these tests were all run with 16 listeners and handlers to keep it short and to the point.
Notice, two of the three tests detailed above are really overdoing it with the server. But, I wanted everyone to see what some bad and worse scenarios would be with the TCP Server Deux component.
And, one more JIC... Yes, the second test listed above had no dropped connections. There is a slight trick to delaying incoming requests when there are no more listeners available. But, I will _not_ go into detailing this other than to say that TCP Server Deux works this way on all incoming requests...
Now, the results... You can view the details yourself at:
http://www.deepskytech.com/tcpsd_tests_20011204/
Let me know reactions to these results. I am quite curious what people thing of these "worst case" propositions with the components.
Steven G. Willis -----------------------------------------
On Dec 4, 2003, at 11:03 PM, Mehboob Alam wrote:
Responses have been pretty slow so far..
for clarification - I am in a strictly Windows environment (at least till 4D actually runs on Linux, but that's another story :)
MacOSX is not an option, hence nor is Webstar, except for my own pet projects.. and I'm too busy for one right now..
For many reasons, I am looking for acquire most of DeepSky's TCP suites and source-code licenses, and I was hoping to hear from more people about their experiences..
thanks, - me _______________________________________________ Active4d-dev mailing list [EMAIL PROTECTED] http://mailman.aparajitaworld.com/mailman/listinfo/active4d-dev
--------------------------------------------------- Flemming Andersen AKTIV Software Corporation [EMAIL PROTECTED] 723 Paskin Way Tel: 250.727.3442 Victoria, BC V8Z 6N4, Canada Fax: 250.727.3740 http://www.aktiv.com
