Re: OO and RDBMS
От | Christopher Browne |
---|---|
Тема | Re: OO and RDBMS |
Дата | |
Msg-id | 60k77x6nyp.fsf@dev6.int.libertyrms.info обсуждение исходный текст |
Ответ на | Re: OO and RDBMS ("Merlin Moncure" <merlin.moncure@rcsonline.com>) |
Список | pgsql-advocacy |
merlin.moncure@rcsonline.com ("Merlin Moncure") writes: > As a follower of the advocacy list, I feel compelled to make a few > observations about the above post. Perhaps I'm speaking to the choir, > but I am passionate about this subject, and I always astounded about the > sheer volumn of misinformation that abounds! I browsed the Purveyor posts on /., and there certainly were some good comments made. There are several drawbacks to the would-be-superior system - It only provides fairly weak transactional "guarantees." Yes, any one transaction can can be rolled back, but it does not appear to extend past that. Which is fine for a single user system, but not for multiuser. - It doesn't do anything about managing domain typing. The increasing support in PG7.4 for CREATE DOMAIN is something that should surely be trumpeted as a strengthening of relational modelling. Purveyor... Doesn't do this... - One of the most significant merits of the relational model is that it allows data access paths that weren't necessarily planned for. If I need to look at all the transaction records generated between March 10th and April 15th, that have some other oddball characteristics, I need only /describe/ the query, and a relational system can do the query. It may not be particularly efficient in the absence of a useful index, but that may well not matter. In contrast, "object systems" are by and large a throwback to the Network model, where queries require writing a program to navigate through the paths provided by the indices. Which is _blazingly_ fast, if the pre-planned paths are the ones you wanted. But if you need something "oddball," the result may be a query program that is both inscrutable and slow. What is unfortunate is that it looks like they have implemented something that's pretty neat that is probably pretty useful for embedding a sort of "database" support into applications in some dynamic languages. But since they _dramatically_ oversell it, apparently because the producers are "youths" that don't have enough perspective to grasp that other systems _do_ have merits, it's likely to get treated as worthless because of the over-sell. -- let name="cbbrowne" and tld="acm.org" in name ^ "@" ^ tld;;
В списке pgsql-advocacy по дате отправления: