WLS 7 Performance vs. WLS 5.1

Question

We're evaluating 7.0 as a upgrade for 5.1.
We have two machines, configured essentially identically.
On one machine, we're running 5.1 sp10 (maybe 9), and the other we're
running 7.0.
We're benchmarking a small part of the system that is "pure" application
server, versus web tier to app server.
WLS 7.0 is clocking in a ~40% slower than 5.1 for this test.
They're running, essentially, the same jar file. For 7.0 we wrapped it into
a simple application EAR file, and we had to make one class change for it to
compile against 7.0.
Other than that, the jar is basically the same, holding 45 beans. They're
all 1.1 Session, BMP, and CMP beans. They both had their respective ejbc
routines run on them, for WLS 7, this was done outside the container.
They are both using different JDBC drivers, the Oracle OCI drivers supplied
with the respective WLS servers.
They're both taking advantage of the internal WLS "pass by reference"
optimization, even though we're not using the LocalInterfaces.
The 7.0 instance looks like it is doing a little bit more GC activity than
the 5.1 instance, but the numbers we're getting back from the
'java -verbose:gc' display aren't enough to come anywhere near the
difference in performance that we're seeing.
So, I'm curious what other sources of internal friction could be?
The test is with ONE user performing, essentially, one transaction. So, it's
a bit of a micro benchmark, but it happens to exercise a majority of the
system. It's not a contention problem.
Maybe it's an unrealistic test. Perhaps that under load with more users,
while WLS 7 is slower now, it's performance will stay more constant as more
load is placed on the system compared to WLS 5.1.
The machines we're testing on are small machines (Sun Netra's, 512MB). The
goal was to get as close to an "apples-to-apples" comparison as we could.
Any pointers would be great.
Regards,
Will Hartung
([email protected])

Answer 1

what security impl are you using for each, they cannot be the same?
Joel
"Will Hartung" <[email protected]> wrote in message
news:[email protected]...
We're evaluating 7.0 as a upgrade for 5.1.
We have two machines, configured essentially identically.
On one machine, we're running 5.1 sp10 (maybe 9), and the other we're
running 7.0.
We're benchmarking a small part of the system that is "pure" application
server, versus web tier to app server.
WLS 7.0 is clocking in a ~40% slower than 5.1 for this test.
They're running, essentially, the same jar file. For 7.0 we wrapped itinto
a simple application EAR file, and we had to make one class change for itto
compile against 7.0.
Other than that, the jar is basically the same, holding 45 beans. They're
all 1.1 Session, BMP, and CMP beans. They both had their respective ejbc
routines run on them, for WLS 7, this was done outside the container.
They are both using different JDBC drivers, the Oracle OCI driverssupplied
with the respective WLS servers.
They're both taking advantage of the internal WLS "pass by reference"
optimization, even though we're not using the LocalInterfaces.
The 7.0 instance looks like it is doing a little bit more GC activity than
the 5.1 instance, but the numbers we're getting back from the
'java -verbose:gc' display aren't enough to come anywhere near the
difference in performance that we're seeing.
So, I'm curious what other sources of internal friction could be?
The test is with ONE user performing, essentially, one transaction. So,it's
a bit of a micro benchmark, but it happens to exercise a majority of the
system. It's not a contention problem.
Maybe it's an unrealistic test. Perhaps that under load with more users,
while WLS 7 is slower now, it's performance will stay more constant asmore
load is placed on the system compared to WLS 5.1.
The machines we're testing on are small machines (Sun Netra's, 512MB). The
goal was to get as close to an "apples-to-apples" comparison as we could.
Any pointers would be great.
Regards,
Will Hartung
([email protected])