Thank you for the reply. I am sorry that what I said is not true. Tested more with version 5.0.28 of Tomcat also produce the same result. Please see if this is "normal" behavior:
I have an axis application, basically return a "hello world", I tried this with:
1) Axis 1.1 2) Axis 1.2 3) Tomcat 5.0.28 4) Tomcat 5.5.4
Some combinations of those. They all produce connection refuse. The max thread is 200, the acceptCount is 100.
On the client side, I also send in some data roughly the size of "hello world".
I have about 30 threads running, each continuously make requests to the server for about 100 requests for each threads. Some tests, I also put random wait time between requests for each thread in around 200 miliseconds. Some test, there is no wait time, after the first request finish, I would make a next one.
After a while, some of the threads will generate exception: connection refused. With 30 threads, this seems to be low to me. Both the client and server sits in a same machime.
My question is this: is this load/scenario normal? If not, I'll be happy to put together a test client and a test Soap Axis server code for you to investigate. You can easily take one of the example client, put it in 30 threads, and spin it 100 times in each thread.
If you have any advice, or suggestion on anything I may have missed, please let me know. Right now, each request seem to have its own connection. Using the common http client for the client to reduce the number of connection, the things is twice as slow, and does not improve the situation (I did not pass any parameters to this connection pool thing, so default value is used on this).
Thank you for your time.
Remy Maucherat wrote:
On Tue, 16 Nov 2004 23:17:22 -0500, V D <[EMAIL PROTECTED]> wrote:
Today, I ran an axis application under both Tomcat 5.028 and 5.5.4.
Hitting this axis application with 100 simultaneous clients for many
hits. 5.0.28 seems to hold up very well. It drops connections once a
while. However, 5.5.4 drops many connections. I wonder if the Tomcat
team aware of this problem.
No. You need to give reproduceable facts, prefereably with a simple application.
Note that the low level code is extremely similar between 5.0 and 5.5. I doubt the few code changes could cause something like that.
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
