Re: Auto Tuning

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

> Nathan,
> It seems you have confused terms here.  Paging
> is not "a useless activity". If you need database
> records, they must be paged into memory. 

The need to move files from disk to memory is obvious.  By paging, I was
referring to the repetitious swapping of pages from memory to disk, caused
by a pool's memory constraint.  That type of workload is something the auto
tuner might reduce, given the chance, by increasing the size of the pool, or
reducing the number of concurrently active threads in the pool.

Similarly, my comments about "manual tuning" were only relative to
suggestions that IBM's auto tuner (QPFRADJ) be disabled, which I generally
disagree with.  I understand needing to manually define time slices, job
priorities, memory pools, subsystems, job queues, job classes, and so forth
which I call workload management, but others might call "tuning".  I
understand the need for workload management, and accept some responsibility
for not expressing myself more clearly.

Discussions about optimizing SQL after the fact, and performance tuning,
sometimes strike me similarly to discussions about treating symptoms of a
disease, rather than eliminating the cause of it.  Hence, my emphasis on
needing to write efficient code and use efficient interfaces, so that down
the road, user's aren't put under pressure to manually "tweak" system
settings, which offer comparatively small gains - often at the expense of
some other job's performance.

I recently tested a couple software products that transformed 5250 data
streams into HTML.  Although functionally similar, one of the products
offered about ten (10) times better performance, CPU wise.  That could mean
the difference of needing a 300 CPW server vs. a 3000 CPW server to get
equivalent performance.

I doubt the more efficient software was more expensive to write, but
depending on how widely distributed, could save a bundle of money overall.

Nathan.