5 ms·

Here here! Well worded! This argument about the relational model have been coming up for years and I see articles like this all the time. The fact is that RD

by Femur 18y ago

Here here! Well worded! This argument about the relational model have been coming up for years and I see articles like this all the time. The fact is that RDBMSs work great when properly implemented and maintained. They are very commonly used and are not going away any time soon.

jrockway 18y ago
The fact is that RDBMSs work great when properly implemented and maintained. Programs written in assembly also work great when they are properly implemented and maintained. The thing is, properly implementing and maintaining it is difficult and time-consuming. Alternatives to assembly made implementing and maintaining software easier; why do you think that this won't happen with databases? (I consider RDBMSs as "assembly" because everyone thinks they can solve every problem. But you have to work at a very low-level to solve certain kinds of problems with a relational database. Something higher-level is an improvement, even if it isn't "necessary". You can write your Web 2.0 app in CPU microcode, but you probably shouldn't.)
- dasil003 18y ago
  That's not really a fair comparison though. There's no parity mismatch between assembly and high level languages. Sure assembly opens a few doors for optimization, but you don't give up any fundamental logic when you move to a higher level language. As for RDBMS, well, they have scaling issues because of the rich functionality they provide, and some queries can definitely be tricky for the same reason. However conceptually they are much more straightforward than assembly, and offer a very pure data model with more overall utility than any of the alternatives. That's not to say that we don't need alternative data stores for some applications. But I do want to go on record as saying this article is a mixture of hype and FUD. RDBMS' have been challenged for decades by alternative supposedly-superior technologies. I think the problem is that no model is as comprehensive as the relational model. It's trivial to write a faster database providing some subset of SQL. You can even reduce the parity mismatch to some languages by making it store objects more directly. That's great, but I have yet to see any serious challenge to the relational model's stengths, instead all I see is hype and tutorials about how to convert some simple query to map-reduce.
  - jrockway 18y ago
    Instead all I see is hype and tutorials about how to convert some simple query to map-reduce. The people that are actually implementing these systems don't have much time to talk about it. I helped write an object database system called KiokuDB almost a year ago, but I still haven't had a chance to write a good blog post about it. There is too much other stuff to do, and writing about databases is an invitation for lots of whiny RDBMS fanbois. (I wrote an article about Elephant, posted it here, and most of the comments whined about how useless object databases are. So whatever, I will use them to do my work, and you guys can discover them for yourselves.)
    - dasil003 18y ago
      Nice indirect ad-hominem. I never said that the relational model is the end-all-be-all for databases. If lumping me into your giant fanboism strawman makes you sleep better then so be it.

by Femur 18y ago

jrockway 18y ago

The fact is that RDBMSs work great when properly implemented and maintained. Programs written in assembly also work great when they are properly implemented and maintained. The thing is, properly implementing and maintaining it is difficult and time-consuming. Alternatives to assembly made implementing and maintaining software easier; why do you think that this won't happen with databases? (I consider RDBMSs as "assembly" because everyone thinks they can solve every problem. But you have to work at a very low-level to solve certain kinds of problems with a relational database. Something higher-level is an improvement, even if it isn't "necessary". You can write your Web 2.0 app in CPU microcode, but you probably shouldn't.)

dasil003 18y ago
That's not really a fair comparison though. There's no parity mismatch between assembly and high level languages. Sure assembly opens a few doors for optimization, but you don't give up any fundamental logic when you move to a higher level language. As for RDBMS, well, they have scaling issues because of the rich functionality they provide, and some queries can definitely be tricky for the same reason. However conceptually they are much more straightforward than assembly, and offer a very pure data model with more overall utility than any of the alternatives. That's not to say that we don't need alternative data stores for some applications. But I do want to go on record as saying this article is a mixture of hype and FUD. RDBMS' have been challenged for decades by alternative supposedly-superior technologies. I think the problem is that no model is as comprehensive as the relational model. It's trivial to write a faster database providing some subset of SQL. You can even reduce the parity mismatch to some languages by making it store objects more directly. That's great, but I have yet to see any serious challenge to the relational model's stengths, instead all I see is hype and tutorials about how to convert some simple query to map-reduce.
- jrockway 18y ago
  Instead all I see is hype and tutorials about how to convert some simple query to map-reduce. The people that are actually implementing these systems don't have much time to talk about it. I helped write an object database system called KiokuDB almost a year ago, but I still haven't had a chance to write a good blog post about it. There is too much other stuff to do, and writing about databases is an invitation for lots of whiny RDBMS fanbois. (I wrote an article about Elephant, posted it here, and most of the comments whined about how useless object databases are. So whatever, I will use them to do my work, and you guys can discover them for yourselves.)
  - dasil003 18y ago
    Nice indirect ad-hominem. I never said that the relational model is the end-all-be-all for databases. If lumping me into your giant fanboism strawman makes you sleep better then so be it.