RE: 4 Disk Raid5 Performance over 3 Disk Raid5 Performance

Rob:

The new system is a 4-LPAR model 520. Not a monster by any means, but pretty 
nice in _our_ environment. Since software development is important to this 
list, it can be significant to listers looking for development systems.

One of the 3-drive LPARs replaces a model 40S that really was used mostly as a 
batch server. It primarily ran compiles, optimizations, etc. It worked 
admirably for a many years, but we reached DASD expansion limits within the 
box. And of course, it was at the end of its OS/400 upgrade path.

The second 3-drive LPAR replaces the model 270 that was specifically for 
development work, i.e., software development, unit testing, initial system 
testing before QA gets involved, etc. Not a bad box at all, but we chose to see 
how the LPAR would work out. There's _no_ way the developers here want to go 
back.

One other LPAR replaces a third machine and a fourth LPAR creates a new system 
image. I'm not sure how many systems we have around here any more, more than a 
dozen counting LPARs as systems. This is our second LPAR machine and it blows 
the previous one away. It's also the first we've ever tried a 3-drive RAID set, 
and there's been absolutely zero indication of anything but major performance 
boosts.

An example previous product build run of one product with full compiles, full 
optimization and all associated steps, would run 18 hours. (Doing optimization 
can be a killer on a *FULL build.) It's dropped to an hour.

I said more than once that our experience is _NOT_ typical. But I can't believe 
we're the only ones on this list that would see similar effects given 
reasonably similar requirements.

Tom Liotta

midrange-l-request@xxxxxxxxxxxx wrote:

  6. RE: 4 Disk Raid5 Performance over 3 Disk Raid5 Performance
     (rob)

I guess if you're upgrading from a B10 to a 520 then a current 3 drive 
raid set won't seem too bad.  However, failing that scenario I sure 
wouldn't recommend a three drive raid set.

midrange-l-request@xxxxxxxxxxxx wrote:

  9. 4 Disk Raid5 Performance over 3 Disk Raid5 Performance
     (kirkg)

I know from real life that a system with a 3 drive raid5 set performs 
almost as bad as a 2 drive mirrored  set. Now I need some official doc 
that says a 4 drive raid5 set out performs a 3 drive raid5  set.  Anyone 
seen something like this?

Unfortunately, no, I haven't seen actual docs. But for the sake of 
completeness on the concept, I want to add a "lunatic fringe" comment... 
NOT to be confused with any form of "recommendation"!

While a 3-drive RAID5 set may perform far worse than 4 or more drives in a 
set, it _might_ not perform badly enough to ignore as a possibility. We 
moved a couple systems from some old hardware to a couple of the 
partitions on a new box, both partitions with 3-drive RAID5 sets, and both 
are so much blazingly faster than the old that we fail to see what the 
complaints are about.

MAJOR caveat -- our processing is ***NOT*** typical and definitely not 
database bound. But when we run a *FULL build of a significant product, we 
still get a big chunk of AuxIO.

The only point is that _sometimes_ for unusual circumstances, a 3-drive 
set can work well. The sum of performance boosts from various components 
after upgrading to new hardware can be impressive.