I suspect that there's a bad problem with thread
synchronization. Has anybody else seen similar symptoms??
Platform: Cyrix M-II running RedHat 6.0
JDK: 1.1.7 from Blackdown
Green threads, default JIT setting (on?)
Classes compiled with JDK 1.1.6 on a WinNT platform
(VisualCafe)
Problem: I've got an application that, among other things,
maintains an "event log". Since the log entries can go to
several places, including but not limited to a local file,
and since I didn't want the logging threads to have to wait,
I put the logging into a separate thread. The interface
between the log-maintenance thread and the rest of the app
is a "bucket" object, which is a limited-size FIFO queue. A
thread can put a log request into the bucket; the
log-maintenance thread extracts the request from the bucket,
processes it, and loops back to get the next request. The
bucket uses synchronized blocks with wait() and notifyAll()
calls to coordinate the maintenance thread with the other
threads. This approach has worked fine on WinNT and Solaris
2.7.
The first thing I noticed when running the app on the
Blackdown JDK was mysterious hangs. To fix this, I first
inserted some yield() calls outside the synchronized blocks,
after each bucket put and get. This helped but didn't
completely solve the problem. I then modified the
maintenance thread so that it runs at a thread priority of
1+whatever the default is -- no more hangs. Hmmm...
Now, the actual environment involves three separate
platforms. Call 'em A, L, and C, where L is the Linux
platform. Platform A sends a request to L via CORBA, which
processes it and sends the result to C. All three platforms
maintain logs of requests. What I'm seeing now is an
occasional double logging on L. That is, A logs requests 1,
2, and 3; C logs results 1, 2, and 3; but L logs processing
1, 2, 2, and 3. Keep in mind that I don't see any such
behavior on either WinNT or Solaris. Hmmm...
I'm concluding that there's something seriously wrong with
thread synchronization, at least on this platform. Proper
synchronization of threads is critical to this application,
and I'm starting to wonder if there are other processing
anomalies on the Linux box that I'm simply not seeing.
Although I haven't made extensive tests, the problem doesn't
appear to be JIT related -- at least, the original app hangs
persisted when I turned the JIT off (java.compiler=NONE).
My questions include:
o Has anybody seen similar synchronization behavior?
o Anybody want to venture an opinion on the likelihood of a
chip-related problem vs a JDK-related one?
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
