Re: Article about supposed "murky" future for Oracle

"Noons" <wizofoz2k_at_yahoo.com.au> wrote in message news:<406ab360$0$15064$afc38c87_at_news.optusnet.com.au>...

> No. Incorrect. And completely ignoring the reality of data processing.
> Versioning has NOTHING to do with isolation levels and it
> solves the problem in a TRANSPARENT fashion.

I agreed with you up until this one, sorry. There are some subtle aspects to Oracle's versioning mechanism that makes its SERIALIZABLE isolation level unlike other databases' SERIALIZABLE isolation levels, so it is most definitely *NOT* transparent.

This is not because Oracle is deficient; it is because ANSI SQL's isolation definitions are ambiguous and inadequate. Oracle's SERIALIAZBLE level is different in that it makes a "non-serial" history possible (this is called "write skew" , i.e. Tom Kyte's example in his expert one-on-one book). This usually is an "OK" thing in practise as long as you're aware of it; the benefits outweigh this anomaly usually. A form of phantom write is also sometimes possible.

This paper is a short and very readable explanation of the problem & how Oracle fits in with its Snapshot Isolation...

http://www.cs.duke.edu/~junyang/courses/cps216-2003-spring/papers/berenson-etal-1995.pdf

Cheers
Stu Received on Wed Mar 31 2004 - 20:01:23 CST