[wellylug]The top 10 dead (or dying) computer skills

[Gerald Roehrbein]

Hello Andrej,

> why bother if you can 
> just throw more hardware at it ;} ... hardware is cheaper
> than programming time.

maybe that's cynism. If not just read that:

I disagree.I'm working since 15 years as a performance expert in huge
SAP and Oracle environments and I've never seen that new hardware solved
performance problem in commercial environments.

In most cases (except correct planned CPU upgrades) new hardware
decreases performance. (Major reason is: Mistakes during upgrade!)

That's not only my experience. Just read Millsap Optimizing Oracle
performance or ask Shallahamer (www.OraPub.com) or read his
publications. I've just published some data gathered by my own at
www.oraforecast.com/CPM (Sorry. My English is really bad but there are
some really good images with statistics I've created which show the
effect of software and hardware tuning).

Most performance problems I've found the last years:

* bad algorithms (
  bad programming in general, 
  missing of not used indexes,
  bad data models,
  bad SQL,
  bad "C","C++","JAVA","ABAP","PL/SQL"
)
* bad configured software 
(databases, 
 application servers, 
 operating systems, 
 filesystems and networks
)
* bad configured hardware
(
 RAID,
 HBA's
)


4GL programmers can not understand most of the reason of bad performing
software. "C" knowledge, knowledge of bits an bytes help.

An absolute must is to understand the complete performance stack
(Goldratt), Queuing Theory (see Wikipedia) and Pareto analysis (80:20
rule) in detail/ reality. 

For example: Most admins know about CPU idle but nearby nobody knows
about CPU run queues and their relations to performance especially in
high parallelized environments.

I've seen for example systems with 16 dual core CPU's with an average
CPU usage of 70% (30% idle) and run queue sizes during this time of 40
and more. 

* Is such a system correct sized?
* Should we buy more and faster CPU's?
* Or should we decrease parallelization to increase performance?


Today I'm in a small performance team in a 270.000 employee company with
dozens of databases up to 3TB and it took 2 years to show all managers
that my ideas solve performance problems and not the ideas of admins
which prefer to buy very expensive hardware.


For gamers and children playing with a PC faster and newer hardware
(CPU's, Graphics adapters or more RAM) will solve performance problems.

This is not valid for commercial applications and commercial systems.

Hardware is expensive. Very expensive. 

* What's the price of 1GB of a high end Hitachi or EMC storage?
* What's the price if you have RAID 1, and one or more Standby Systems?
* What's the price if you have a hardware for stage development, test
and production?
* Maybe you have 1GB "mirrored" 6 or seven times. It's maybe 7GB for 60
NZD per GB.
* What's the price of 1GB RAM in this case? A high end SUN or Fujitsu
SPARC server takes more as 100NZD for 1GB.

* What's the price for downtime or activating/ deactivating standby
systems (including the risk it may have)?

In commercial environments hardware upgrades will take a lot of
resources (man power) and a lot more as the price of one CPU or 1GB disk
space or RAM.

Wasting resources with bad performing software or inefficient usage of
memory, CPU, storage or Network Bandwidth is expensive.




kind regards
Gerald


> 
> 
> > Cheers,
> >
> > Cliff
> Cheers,
> Andrej
> 
> 
> -- 
> A: because it messes up threading
> Q: why should I not reply by top-posting?
> A: No.
> Q: Should I include quotations after my reply?
> 
>