Re: What affects performance?

In article <CqsC5J.H96_at_rosebud.strinc.com> amw_at_rosebud.strinc.com (Allen M. White) writes:
>In article <2sk35o$80v_at_goshen.connected.com>,
>Kurt Schmidt <slores_at_goshen.connected.com> wrote:
>>In article <Cqq0no.9x9_at_boi.hp.com>, Jack Parker <jparker_at_boi.hp.com> wrote:
>>>
>>>But more seriously, Paul and the other posters made some very valid points.
>>>What the original requestor was asking for was a white paper on performance,
>>>or rather the basis to write one.
>>>...
>>>For example: What's
>>>faster a Maserati or a schoolbus? But are you trying to win the Indy 500
>>>or move people from point a to point b? Sure the Maserati can get there 3
>>>times faster but it can only carry two people.
>>>
>>>I am now supposed to read a white paper which states this? Sure it's possible
>>>to document in general what makes a system fast. But how high is the sky?
>>[stuff deleted]
>>
>>I think it would be quite worthwhile to describe, or to try to describe, various
>>parameters that influence the performance of a system. Or in the words of
>>the example above, mentioning to the reader that the number of people
>>being transported must be taken into consideration, as well as the speed
>>priority. Other items could include information on how to analyze the data
>>being stored, in order to make a reasonable decision (ex.: if BLOBs are
>>involved, what would be of importance? If BLOBs are NOT involved, would the
>>same parameters need to be considered?). Sky's pretty high, but we did touch
>>it, and went even farther :-).
>
>Why would it seem reasonable for us to discuss the specific effects on
>performance here when Sybase offers a 5-day class on the subject? Things
>that directly affect performance are:
>
>1) Volume of data
>2) Clustered indexes
>3) Non-clustered indexes
>4) Raw partitions (vs. data files)
>5) Memory
>6) Number of users, both using the database and others
>7) CPU size and speed
>8) Parameter settings
>9) Network performance
>
>and most importantly
>
>10) How the SQL code is written (with respect to the optimizer)

Wrong! Most importantly is:
10) Database design

The logical and physical design of a relational database is the single most important factor that affects performance. You can tune all the other stuff all day, and if the database design is lously, you'll get lousy performance. I do agree, though, that how the SQL is written is on the list, along with the other things mentioned. I do agree also, however, that this forum is somewhat inefficient for a serious discussion of performance. An important indirect factor, and in fact one of the more common mistakes made by organizations new to relational databases, is the resistance to training. Proper training in a particular RDBMS, in particular performance training, is crucial to the successful use of that RDBMS. I don't say this because our company conducts training classes. I say this from years of experience seeing companies trying to learn INGRES with JOT. While you certainly can build applications without formal training, I typically see those initial applications rewritten once the mistakes have been made. While forgoing training may be cost effective in the short term, in the long run it would have been much cheaper to invest in the excellent and relatively inexpensive training that is available.

>
>ALL of these things have to be taken into consideration when considering
>performance. And I'm sure I left out a whole bunch of other considerations
>as well, so basically you should take a WAG and go with it.
>
>-----------------------------------------------------------------------------
>Allen M. White | To run is to live o'
>STR, Inc. | To live is to run <|v
>6800-A West Snowville Road | I'm running for my life! />
>Brecksville, Ohio 44141-3214 USA |
>(216)546-9512 x227 | CWRRC - Cleveland West Road Runners
>amw_at_strinc.com |
>-----------------------------------------------------------------------------

-- 

Steve Caswell           |   (404) 448-7727    |  "The opinions expressed are my
Principal Consultant    |   sfc_at_tpghq.com     |   own.  They may not be perfect,
The Palmer Group        |   uunet!tpghq!sfc   |   but they're all I've got."