-----------------------------
Please read the FAQ!
<http://java.apache.org/faq/>
-----------------------------
<CONFIG>
Apache JServ: 1.0b5
Apache: 1.3.4.
JDK: 1.1.7
JSDK: 2.0
OS: Solaris 2.6
DBMS: Oracle 8.0.4.
<\CONFIG>
We're having very serious performance problems with the servlets we have
developed. They run reports against and Oracle database, returning the
results to the clients' browers. The reports take a fair amount of time (as
much as 1 minute) to run since they contain a lot of data. The real problem
is that two reports triggered simultaneously run approximately twice as
slowly as one, three reports three times as slowly etc.
We have identified a number of possible causes of this scalability problem
all of which we have subsequently discounted:
1. Database locking: some of the reports are read-only and as such should
not be locking the tables they access. Oracle profiling tools have indicated
that no locks are being applied.
2. Database connections: again our Oracle profiling tools suggest that all
the connections in our connection pool are being used more or less equally
with no blocking. Furthermore we have written some test servlets which don't
do any database access and the same problem still affects them.
3. Deadlocking in our Java code: the OptimizeIt Java profiling tool
indicates that although the total time to run reports increases as the
number of clients goes up, all our classes and methods are taking the same
proportions of the total time. OptimizeIt should show up any bottlenecks
which multi-client access causes but it doesn't seem to show any.
We are not running any of our servlets with the SingleThreadModel. If we
run two different reports (i.e. two different servlet classes) at the same
time they both take about twice as usual.
4. System resources: since the reports are quite large, and involve a large
number of objects being created on the server I have increased the starting
heap size available to our JVM to 150M and the maximum heap size to 300M.
This hasn't made any difference to our problem.
As you can see we've tried to come at this problem in a number of ways with
no success. Can anyone suggest any possible solutions? We're pretty well out
of ideas ourselves!
--
Noel McNulty
Belfast Engineering Centre
Information Systems Engineering
BT
--
--------------------------------------------------------------
Please read the FAQ! <http://java.apache.org/faq/>
To subscribe: [EMAIL PROTECTED]
To unsubscribe: [EMAIL PROTECTED]
Archives and Other: <http://java.apache.org/main/mail.html>
Problems?: [EMAIL PROTECTED]
