Re: Hierarchical Relationship

From: Jan-Erik Rosinowski <spamfilter_at_rosinowski.de>
Date: 2000/03/19
Message-ID: <38d4b3ee.2349308_at_News.CIS.DFN.DE>#1/1


joe_celko_at_my-deja.com wrote:

>>> we use prefixes for tables (t_), views (v_), functions (f_) and
>procedures (p_). in case of performance problems one easily sees
>what's going on. <<
 

>No, you don't see anything about performance from the prefixes. I owld
>guess that 80-90% of the time, bad performance is due to a bad query,
>not to how the data was physically stored.

ok, my explaination was rather brief: consider team-development, you have to get a quick overview of "where you are" and what possible implications your changes will have. this prefixing gives you a first and imho valuable insight. can i make changes to this relation or is it a view which is not updateable etc..

>>> ever wondered how a query like "select a from b" could take ages?
>let "a" be a function without arguments... sure, the compiler knows the
>types and semantics but black-box-thinking and information hiding works
>only to a limited extend. <<
>
>So much for the whole ideas of Structured Programming, OO programming,
>and most of computer science for the last 25 years!<g>

i remember lectures about compiler construction where some students really considered implementing symboltablemanagement thru a database <bg> do you think this is reasonable?

>I do not do
>host language operations in my database, such as elaborate user defined
>functions. We designed SQL to be a data retrieval langauge, not a
>computational language.

there are lots of scenarios where it makes sense to do the computing on the serverside. implementing aggregate functions on the clientside is a rather suboptimal idea.

>When I do get to the point that I know I have a good query, but a poor

how do you ensure to have a good query when the underlaying optimizer is "cost-based" and statistics vary?

am i the only one who is sick of this stuff and who is wondering why there are still "obsolete" rule-based optimizers used in mission critical environments? not sure but i was told that even oracle financials does this :-) you think we should completely ignore whether we'rre dealing with views or tables? hmm

>execution, I look at the execution plan (EXPLAIN , SHOW, whatever that
>particular product uses for a keyword). Now it is a matter of
>statistics and indexing, not whether something is a table or a view.

i've studied informatics / computer science and i'm working in this field for more than 15 years. i think i know the theory but i also see the problems in practice

i hope my point got a little clearer

ciao, jan

http://www.rsp.de/

rs&p-Dossier: Software zur Erstellung technischer Dokumentationen

              und Schriftgutes in Verwaltung und Industrie. Received on Sun Mar 19 2000 - 00:00:00 CET

Original text of this message