RE: DDL vs. DDS performance

["Joe Pluta" <joepluta@xxxxxxxxxxxxxxxxx>]

> From: Elvis Budimlic
> 
> In your testing, do you purge the objects out of memory between the
runs?
> I think the best way to accomplish this is via SETOBJACC POOL(*PURGE)
or
> alternatively CLRPOOL commands.
> If you're not purging the files out of main memory, I can see where
your
> test results would differ from Dan's.

But that in turn begs the question: is clearing the memory pool really
representative of normal load?  In my thinking, the beauty of OS/400 is
that it does all this wonderful caching for us, and that's why the box
shines as you put more people on it.  I never clear pools in production.

By clearing the pool every time, you're in effect setting up a
controlled but unrealistic test environment, at which point it is harder
to compare any real gains versus the losses, for example, from access
path sharing (or lack thereof).

I'm a big proponent of benchmarks, but I always have to remind myself
that all a benchmark gives me is the results under specific
circumstances.

Joe