New advanced book on index design

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

Lauri - 21 Aug 2005 22:15 GMT

There is a new book on indexing SQL-DBMS'es
by Tapio Lahdenmaki and Michael Leach.

Relational Database Index Design and the Optimizers
See
http://tinyurl.com/84g95
(Amazon-link)

It is probably a bit heavy for beginners
but experienced DBA's should find this
book very usefull.

Tapio Lahdenmäki is my performance "guru"
and I have learned maybe 80% of what I know
about DBMS-performance from him.

regards,
Lauri Pietarinen

Reply to this Message

BobTheDataBaseBoy - 22 Aug 2005 04:27 GMT

> There is a new book on indexing SQL-DBMS'es
> by Tapio Lahdenmaki and Michael Leach.
[quoted text clipped - 14 lines]
> regards,
> Lauri Pietarinen

saw the title, intrigued (and get a massive discount since The Wife
works at a bookstore), got it.

couple of points i don't like:

the authors treat a RDBMS from the point of view of COBOL cursor
traversal.

they don't spend any time discussing how to force buffer writes to
commits; the source of real performance (a bit on the edge, sure).

there's no math of significance.

the Tow or Gulutzan books provide (mutually different) points of view on
performance. you really need all three. not to mention your vendor
manuals.

BTDB

Reply to this Message

Lauri - 22 Aug 2005 07:12 GMT

>> There is a new book on indexing SQL-DBMS'es
>> by Tapio Lahdenmaki and Michael Leach.
[quoted text clipped - 6 lines]
> saw the title, intrigued (and get a massive discount since The Wife
> works at a bookstore), got it.

Wow, that was fast!

> couple of points i don't like:
>
> the authors treat a RDBMS from the point of view of COBOL cursor traversal.

That is probably true since their background is in the 70's and 80's and
in big iron MVS/DB2 -systems. Still, I would argue that most issues from
those days are still valid.

> they don't spend any time discussing how to force buffer writes to
> commits; the source of real performance (a bit on the edge, sure).

How does that happen? And why would you want to force buffer writes
to commits?

> there's no math of significance.

In the 80's his courses (Lahdenmaki) had lot's of math, but it turned
out that the formulas were not very usefull in practice. The
formulas where simplified ever more
(see e.g. http://www.redbooks.ibm.com/redbooks/pdfs/sg242549.pdf) and
his latest courses have further deemphasized exact calculations.

What particular math were you missing?

> the Tow or Gulutzan books provide (mutually different) points of view on
> performance. you really need all three. not to mention your vendor
> manuals.

I have not read the Tow or Guluzan books. What is the viewpoint that
the new book does not cover (and which one does it cover well, in your
opinion)?

One thing that I personally think that is missing is the application
architecture view point, i.e. how architecture influences performance.
The viewpoint of the book is strictly confined to the DBMS, i.e.
how to get stuff out of the DBMS as fast as possible.

However, the title is "Relational *Index* Design and the Optimizers"

Regards,
Lauri Pietarinen

Reply to this Message