Are there any numbers for comparing TCP and BEQ connection performances when used with baan and oracle at the latest released versions.

We are evaluating the moving from a host mode installation to a database server and an application server.

Thanks,
Giuseppe

No, there are no figures but it all depends.

How much depends on a few things like:

o CPU speed.
o Network cards
o OS. Some OS have a faster network stack than others
o Amount of bytes and packages to be sent across the network
o Used database
o Database configuration

In general it can be said you loose at least 15% performance. As Baan is doing lots of small requests, it doesn’t matter that much if you do batch or OLTP.

Hope this helps,
Dick

Well, of course it depends from hardware and software, but I was wondering if any general data was available.

>o CPU speed.
5 CPUs POWER5, 1.5Ghz on DB server
4 CPUs POWER5, 1.5Ghz on APPL server

>o Network cards
unknown, but we may change them

>o OS. Some OS have a faster network stack than others
AIX 5.2

>o Amount of bytes and packages to be sent across the network
just ora8_srv6.1 -> oracleSID connection, in a "normal" BAAN IVc4 installation

>o Used database
Oracle 9.2.0.4

>o Database configuration
We may change the configuration, if required.

Thanks,
Giuseppe

We did a test a couple of weeks ago.

C/S Combo Mode Additional Time
Y N 1h30m
Y Y 1h25m
N Y 1h
N N 1h10m
N Y HT off 59m

The test was done by running a costprice calculation for about 70k items (single items, no BOM).

Regards,
Han

Thanks for your data. You write an additiona time absolute. Do you alsa have percentages? What are the timings in host mode?

We did a test a couple of weeks ago.


Hi Han,
I finally understand your message (it was screwed up by the html layout)
so you are basically writing that total time changed from 1h10m to 1h30m
or from 1h to 1h25m, depending from the 'combo mode'.

I have another question: did you tried this on a low loaded machine?

I believe that on a heavy loaded system, the total time in 'host mode' is
actually more than this, because we are missing CPU power (the run queue
is about 2-3 times the number of available CPUs), so we may expect as
"worst case" your numbers.

Sorry for the messy table.

The number that you mention are all compared to using client/server or not.

It was a new server, not loaded at all.

Your runqueue numbers sounds as if the machine is cpu bound.

Regards,
Han

I have another question: did you tried this on a low loaded machine?
I expect that combo will perform even better on a loaded system.

I believe that on a heavy loaded system, the total time in 'host mode' is
actually more than this, because we are missing CPU power (the run queue
is about 2-3 times the number of available CPUs), so we may expect as
"worst case" your numbers.
If your system is heavy loaded, You always win to move to client server mode (if the CPU's or network are not too bad). So in your case I advise to move to client server mode. A run queue of 3 times the number of CPU's becomes bad. Moving a part of the load to an application server will most probably increase performance instead of a decrease as mentioned earlier.

Kind regards,
Dick

Right, this is what I suggested too.

Thank you very much.