In general the entity engine performance test page in webtools is not
a good predictor for web site performance because it is not really
similar in any way to what happens in real-world use of an
application. It only runs in a single thread (so multiple processors
won't help much), it does a bunch of pretty bogus, repetitive data
operations with no presentation, which is also very different from
real application usage.
The best thing to do is have a separate box (or a number of separate
boxes...) run something like JMeter to do page requests and sessions
that are similar to real world usage. There are various open source
and commercial tools for this and for any performance test this is
the only way to go.
If you plan on having any sort of load on your servers doing this
sort of thing should be considered necessary. If you find performance
problems explaining and fixing them requires a lot of knowledge,
basically about every piece of software and hardware involved. With a
bit of optimization you can often dramatically improve results though...
-David
On Sep 11, 2006, at 6:33 AM, Fabian Gorsler wrote:
Hi all,
we are currently testing our OFBiz-project on several platforms. The
Entity Engine Performance Test from Webtools is up to now the only
performance test we're doing, everything else is measured "how it
feels".
;)
OFBiz runs fine with all platforms (Windows XP, Linux i386, Linux
AMD64)
with J2SE 1.5 from Sun. We could not encounter any platform-specific
problems. :) This evening I will test running OFBiz on Solaris 10 with
J2SE 1.5 again, but I don't think there will be any problem. Pure Java
rocks. :)
BTW: Has anybody any ideas what perfomance tests could else be
done? We
want to stress OFBiz and I think the best ideas for stressing have the
developers. :)
So, back to the problem: We have OFBiz running on an two-way
Opteron 265
(Dual-core, there are four physical processors) with 4 GB of RAM, the
operating system is Linux (2.6.8). PostgreSQL 8.1 runs on this
machine as
our database. All works fine, but the Entity Engine Performance
Test shows
us poor results compared to other systems we have tested. Running
the test
on a Windows XP-box (2.8 GHz Celeron, 2 GB RAM) with PostgreSQL on the
same machine is much faster. Those results are shoking us. We
thought the
Opteron-machine would blame the other hosts...
The configuration of the test scenarios in detail (in all scenarios a
version of OFBiz from trunk was used):
Test 1:
Opteron 265 (Dual-Core, two-way, four physical CPUs), 4 GB RAM,
U320-SCSI
RAID1, Debian Sarge (Linux 2.6.8, i386), PostgreSQL 8.1 (on the same
host), Java 1.5.0_06
Results: http://www.der-moloch.de/files/ofbiz-ml/
entityengine_opteron.png
Test 2:
2.8 GHz Celeron, 2 GB RAM, single IDE-drive, Windows XP, PostgreSQL
8.1
(on the same host), Java 1.5.0_06
Results: http://www.der-moloch.de/files/ofbiz-ml/entityengine-
windows.png
Test 3:
OFBiz on an Athlon X2 4200+ (two physical cpus), 2 GB RAM, single
SATA-drive, Debian Sid (Linux 2.6.17.13, AMD64), PostgreSQL 8.1 (see
below), Java 1.5.0_08
PostgreSQL 8.1 on a Sun Ultra 5 (USIIi 400 MHz), 512 MB RAM, single
UW-SCSI-drive, Ubuntu Dapper (Linux 2.6.17.11)
Results: http://www.der-moloch.de/files/ofbiz-ml/
entityengine_ultra5.png
On all machines there were no bigger jobs during the tests and no
other
users were using OFBiz. On all machines the except the Windows host
build
405 and 407 of the PostgreSQL JDBC3-driver were tested (with the same
results). We have even tried to optimized the PostgreSQL
configuration on
the Opteron but this had no effect. All other test scenarios blame our
Opteron...
We think the problem is not PostgreSQL, because we've tried different
configuration options and the postmaster processes on the Opteron
are just
idling around. Has anybody an idea what we could do? Do you need more
information?
TIA
Best regards,
Fabian.
