zillablog - Comments

Robert Young: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Young) — Tue, 09 Mar 2010 19:39:04 -0500

I'm way too lazy to type it all, and it's not on the web, so far as I know, but here's the bottom of pg. 485:

So ACID is a nice acronym - but do the concepts it represents really stand up to close examination? ... in general, no.

His objections aren't just syntactical, although they are somewhat picky, in my opinion.

Robert Young: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Young) — Tue, 09 Mar 2010 19:31:09 -0500

Jeff: >> That might be an incidental benefit.

Robert: >> but the freebie that comes along for the ride is, loosely speaking, a minimal cover of the datastore.

OK, so I got a little carried away. OTOH, one might argue that parsimony of data is a logical consequence of the RM. I know many writers (I'd need to do some searching for a linkable quote; most of what I know is from dead tree sources) have made the point.

Jeff Davis: Actually, the Relational Model doesn't scale

nospam@example.com (Jeff Davis) — Tue, 09 Mar 2010 16:06:18 -0500

"The point of the relational model... is to slay the scalability dragon by avoiding the byte bloat in the first place."

Surely that's not the point of the relational model. That might be an incidental benefit.

Darren Duncan: Actually, the Relational Model doesn't scale

nospam@example.com (Darren Duncan) — Tue, 09 Mar 2010 16:00:22 -0500

In regards to Date's "dropping ACID", what little I know about it (second hand) says that the proposal is to just drop the exact terminology, not the principles behind them; AFAIK, it is more of a subtle thing, arguing about terminology minutae, and is nowhere near as dire as the title "dropping ACID" may suggest.

Robert Young: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Young) — Tue, 09 Mar 2010 15:34:08 -0500

Finally. The point of the relational model (and industrial strength SQL databases) is to slay the scalability dragon by avoiding the byte bloat in the first place. BCNF databases are the most parsimonious storage; removing redundant data from both the model and the implementation is most of the point of the RM. The reasons for doing this, from Codd, were related to data integrity (removing update anomalies), but the freebie that comes along for the ride is, loosely speaking, a minimal cover of the datastore. With SSD and multi-core/processor machines, there is no join penalty, and thus no reason not to build with BCNF databases.

But we're not there yet, in large measure because the OO folk think xml is something really neat and new. Sigh.

Robert Young: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Young) — Tue, 09 Mar 2010 15:28:40 -0500

>> (BTW, getting people to admit we can throw away ACID is an accomplishment in and of itself imo)

Rule 5: The comprehensive data sublanguage rule: The system must support at least one relational language that
1. Has a linear syntax
2. Can be used both interactively and within application programs,
3. Supports data definition operations (including view definitions), data manipulation operations (update as well as retrieval), security and integrity constraints, and transaction management operations (begin, commit, and rollback).

It was Gray and Reuter who wrote the first definitive text on transactions, and made ACID a term that didn't mean LSD. However, Rule 5 is implicitly ACID, otherwise it is meaningless. I will note that Date, in his latest edition, jumps the shark with a subsection "Dropping ACID". He's wrong, too.

Robert Young: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Young) — Tue, 09 Mar 2010 15:19:09 -0500

>> Note that foreign key constraints and constraints themselves, again, are not instrinsic to the relational data model; a nice thing to have, but not required.

Rule 10: Integrity independence: Integrity constraints must be specified separately from application programs and stored in the catalog. It must be possible to change such constraints as and when appropriate without unnecessarily affecting existing applications.

So, wrong.

Robert Young: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Young) — Tue, 09 Mar 2010 15:08:24 -0500

>> Imagine a RDBMS that offers no constraints support and no transaction support as we know it

It's called MySql < 5.x, and has no usefulness other than what it is: a SQL parser fronting the file system, which allows coders who use it to claim that they "do relational", while doing exactly what a 1970's COBOL coder did with VSAM (less, actually).

Jeff Davis: Actually, the Relational Model doesn't scale

nospam@example.com (Jeff Davis) — Tue, 09 Mar 2010 13:06:32 -0500

"getting people to admit we can throw away ACID is an accomplishment in and of itself"

["throw away" is too strong... I think you mean "make optional" or "make not sacred"]

Exactly. That's the point.

As long as people think that the relational model is the problem, nobody (particularly not RDBMS people) will actually fix the real problems. Additionally, it will make people think that the scalability problems are inherent, so they turn to much weaker languages.

That's why, whenever someone is talking about scalability and the relational model, we need to set the terminology straight. If you read my post on "terminology confusion" (cited in my post), you will see how much time is wasted when two sides of a debate are basing their arguments on different definitions.

Robert Treat: Actually, the Relational Model doesn't scale

nospam@example.com (Robert Treat) — Mon, 08 Mar 2010 23:20:30 -0500

Yeah, I know the title of this post isn't fair (which is why I tried to qualify it). The problem is that defending the relational model in a discussion about scalability misses the point; how do you get a all these relational systems we so love to actually friggin' scale? I think that is where the RDBMS folks need to spend their time; not in explaining why the relational model is above the discussion. (BTW, getting people to admit we can throw away ACID is an accomplishment in and of itself imo)