Running the attached program on sparc shows a major performance bug. Running on
Windows NT is smokin fast. The JDK in question is 1.3.1 and 1.4 latest bits.
Looks like a libraries issue in net socket io..

Results are as follows for Compiler 1 & Compiler 2:

With a call to mOutputStream.reset(); in client and server programs
platform Time Iterations for Client - Server
Sparc 38 mins 10
Intel 1 1/2 mins 10

Without a call to mOutputStream.reset(); in client and server programs
platform Time Iterations for Client - Server
Sparc 1 min 10
Intel 45 secs 10


Download Attachments: Then follow steps below.
Steps to reproduce:
1. in one window: java HotSpotFailureServerMain
2. in another window: java HotSpotFailureClientMain
3. Start your Timers..
gary.collins@East 2000-12-14

If you run the program on a pc you will see it make 10 * 4 server transactions with the client execute in 1 1/2 minutes. Running on a sparc it takes 38 minutes +-

I think the issue is with writing an object across a socket, and after it finishes writing the tester code flushes it and then makes a call to reset() method for the Output stream. The reset() method on sparc is causing a major performance issue.