DB Superstar Jim Starkley touts NuoDBs new patent
gigaom.com
gigaom.com
Here is the actual patent: http://www.google.com/patents/US20110231447
I'll write a longer reply when I have time to digest the legalese.
That said, Jim Starkley seems to be pretty patent-happy. Check out this one from 2008, "Method and apparatus for generating web pages from templates" [1].
After all of that, I am unable to discern the novelty, having designed and implemented quite a few distributed database systems. The patent describes boringly standard internals that can be found in many existing distributed databases. In fact, I stood up a system on an EC2 cluster a couple years ago that had remarkably similar internals to what is described in the patent. Maybe I am missing something but I am skeptical.
The design described in the patent still has the same problems and limitations that have plagued distributed OLTP systems for years. The patent appears to be a minor rearrangement of standard architectural elements common to a lot of distributed systems. What is described will work but I could buy equivalent functionality from other vendors.
I was hoping for an interesting computer science-y technology but that patent doesn't contain one. It appears to be a description of a particular implementation of a conventional distributed transaction processing system. What am I missing?
I've been waiting for something like this. NoSQL products took a look at the limitations of SQL DBMS products and (incorrectly) assumed that SQL was the problem. SQL may be a symptom of the problem, but the deeper problem is conflating the logical and physical data model in SQL-based DBMS and also in most NoSQL products. A "record" (however you define it) doesn't need to be mapped 1-to-1 with the physical bits on storage, but for some reason (probably because it's easier) virtually every product does so.
It doesn't matter what language you use to access your data, as long as the underlying DBMS operates on solid fundamental principles of data management (ACID, logical/physical orthogonality), you can have an innovative, pleasing-to-use system.
Very interested to see what these folks come up with.
However, decoupling the problem this way without true ACID transctions is very difficult. I think this is why there are so many NoSQL databases in the market with so many different data models; without transactions a change to a logical "record" cannot atomically update multiple physical "records" atomically.
What has traditionally separated the NoSQL databases from the SQL databases is that they implement a smaller subset of relational algebra e.g. the famous lack of join operators. This gave them the latitude to throw out the SQL interface and build interfaces that were more domain specific and a better match for what some applications need to do.
Now we are starting to see new databases that are discarding the SQL orthodoxy but with a relational algebra implementation that is as expressive as anything found underneath a conventional SQL database. They could implement SQL if they wanted to but there is less of a compulsion to do so than there used to be. Hopefully this will lead to the best of both worlds.
The Relational Model is (hugely simplified) predicate logic and set theory combined with logical/physical data independence and integrity.
Relational algebra (selection, projection, union, etc.) are important but just a part (very important, of course, as it gives us a sound theoretical foundation to implement from) of the query language. The "Management System" part of the "DataBase Management System" are the constraints, logical/physical representation, data model, etc. - which IMHO represent the bulk of what a DBMS ought to do.
What sounds interesting about Nuodb is that they've kept the sound relational fundamentals and created something that is what the relational model has espoused from the beginning. Or so it reads.
Regardless of any innovation NuoDB may bring, can this phrase really be credibly stated, given everything we've seen in the patent arena?
Edit: and the article ends with this paragraph: "We all know technology patents are tricky business — see GigaOM’s Jeff John Roberts’ continuing coverage of the raging patent troll epidemic — but it does show that NuoDB may really be onto something here with its self-scaling elastic database."
However, no technical details are given as to what the innovations are, just that the underpinning technology is so innovative that there weren't even names for what they're doing.