[rescue] Anyone interested in a Sun E3000?

[Jonathan C. Patschke]

On Thu, 30 Nov 2006, Francois Dion wrote:

> That is crazy... what is it doing, ERP for IBM worldwide??

It could just be poor[0] database usage, as well.

$ork[-1] had a cluster of two IBM 7026-6H1 (pSeries 660) systems full-up
of CPU and healthily outfitted with memory solely devoted to running one
Oracle database.  This database wasn't that large (5GB or so) but had a
very bad design.  Some tables had -hundreds- of columns due to very poor
normalization.  Each table had roughly 100,000 rows.  I disremember the
exact number, as I've had an enitre year to put the trainwreck out of my
mind.

Due to the overuse of normalization in some parts of the DB and the
utter -lack- of it in other parts, performance sucked as a baseline.
They also refused to use indexes, since "there has to be a trade-off, if
one type of query is going to be faster, everything else will have to be
much slower!"

Then there was the application that ran on top of it.  It didn't use
stored procedures.  It didn't use prepared statements.  It didn't cache
information.  It didn't use persistent connections.  They weren't even
optimised creatively; they were machine-generated from a pointy-clicky
SQL widget under the dictate of "our programmers must not be required to
know any programming language to contribute to the project."

I was appalled at how slowly our pair of p660s ran that DB.  I could use
the web frontend to our application and, by performing the absolute most
common search, bog down an entire CPU on one of the p660s for -twenty
five seconds-.  If I clicked the "show next 10 results" link, another
twenty-five seconds, and so on.

The best[1] money they ever spent on that project was paying an actual
DBA to come onsite and tear apart the SQL statements, the DB cluster,
the DB schema, and implement it in a way the code-generator just
couldn't.  CPU time for most queries dropped to just under 0.1% of the
original runtime, and it could easily be proved that the difference
would've been more dramatic, had we waited until we loaded the rest of
the data from the mainframe.

If we'd have done that from the beginning, a pair of 7028-6C1 (pSeries
610) systems would've been more than sufficient for the DB load.


[0] However, given that this is IBM we're talking about, I'd hope this
     isn't the case.  Although the horrible system I'm describing in this
     message -was- built by IBM Global Services....
[1] Although, really, that's not saying much, in light of the REALLY
     STUPID things they threw money at.
-- 
Jonathan Patschke  ) "Some people grow out of the petty theft of
Elgin, TX         (   childhood.  Others grow up to be CEOs and
USA                )  politicians."              --Forrest Black