RE: What is Performance

["Nathan M. Andelin" <nandel@xxxxxxxxxxxxxxxxxxx>]

> Dieter wrote:

> If you get 10 to 100 times (better performance
> in RPG), there must be a problem with the design
> of the java code.

Or, could it be that Java and J2EE interfaces run under constraints that
aren't applicable under ILE interfaces?

I've written about J2EE constraints in other settings, and wouldn't care to
rehash them again here.  There comes a point, as Joe Pluta suggested, where
we need to move past theories, and prove it.

If we were to come up with realistic benchmarks, I may be able to contribute
a 100% ILE implementation, which could be compared to someone's 100% Java
implementation, which might be compared with a mixture of Java and RPG.
We'd need a volunteer to host and stress test each implementation on their
server.

IBM published a Redpiece about Intentia's Servlet based order entry
benchmark, running on a model 840-2420, balancing the load against 5
instances of Websphere, using 100-150 threads per Websphere instance, which
pooped out at 50,000 "transactions" per hour.

Interestingly, IBM also published the more rigorous TPC benchmark, written
in COBOL and C, also running on a model 840-2420, completing and journaling
over 9,000,000 transactions per hour.

Were the TPC transactions more trivial than the Java transactions?
Actually, reading both publications in detail, gives the impression that the
TPC transactions were more complex, using more files, larger files, and more
complex record formats.  But it's hard to know for sure, because so many
details about the Servlet benchmark were left out.

Anyway, it would be good to publish some realistic "apple to apple"
comparisons.

Nathan.