Is the NoSQL Meeting Announcing the End of the RDBMS Era?

I read this as "It isn't the right tool for our job, so it is dead". When are we going to realize we can't have a toolbox full of hammers?

I disagree with the RDBMS can scale just as well quote in the article. To take a couple of mature examples, it is very clear that Google Bigtable and Amazon Dynamo scale better than RDBMSes. Of course, they also do less -- that is why. Right tool for the right job...
Dwight/mongodb.org

It's obvious that RDBMSes are definitely the main keepers of our data after 'filesystems', and are likely to be for many years to come, although it's certainly true that the NoSQL approaches are gaining some momentum, particularly for the extra-large web applications.

NoSQL I agree, SQL is full of flaws, but I do not think that the end of RDBMS is near, in fact, they are just beginning.


I guess one of the problems is that since most of currently pseudo relational databases use SQL, people just believes that SQL = Relational, and that is just not true.


All this non-relational databases sound nice (COBOL sounded nice to a while ago) and they sure seem to be good at storing a huge amount of data, but, what about actually working with that data in a consistent predictable way? (As in if this business processes is not done accurately the government/customers will sue us to bankruptcy/jail).


I mean, for heuristic stuff, this non relational databases are fine, but if you really need control of the data, you need a RDMBS

This is kind of silly. RDBMD will be around for a long time to come. This technology is interesting in that it's being used for places where the databases don't work for cost/scale or availability reasons and also to front existing databases for customers with smaller requirements. I think the growth area for this scale out dbms stuff is warehousing. I can see hadoop type systems being used as the primary dataset for warehouse datasets and have it push snapshots or processed sets to conventional warehouses where this makes sense.

He's not saying that Relational databases (RDBMS) are dead/dying. I think he was saying that the era of using one-size-fits-all RDBMS (like Oracle) for all database applications was ending and that a new era of specialized DBMS usage (for data warehousing, stream processing, text search, scientific workloads) would emerge.

In fact, he's still inventing SQL RDBMSes. Vertica is a specialized MPP/columnar SQL RDBMS for handling analytic/data warehousing workloads. VoltDB is a new SQL OLTP RDBMS with a new way of guaranteeing transactional ACIDity for large-scale, clustered, high-transaction volume workloads (versus the slow/non-scaling ACIDity approaches of the 30-year-old OLTP RDBMSs). Disclaimer: I work for VoltDB.

Off the top of my head, I know less about the scientific DBMS he's invented SciDB. I don't believe it's relational though. It's open source and worth reading about; lots of cool design concepts in it.

Thanks,

Andy

Hi Billy,
Most of the article contains what others think on the issue.
My closing statement was: "It is completely unlikely relational DBs to disappear any time soon, but it is possible to see a gradual move towards open source non-SQL data stores for costs, simplicity and scalability reasons."

Hi Andy,
it may be possible I misunderstood M.S. After reading his article in the ACM Blog, entitled The End of a DBMS Era (Might be Upon Us), my impression was that he predicts Relational DBMS will be replaced by other DBMSes: column, document, etc. And that's my article about. The only twist is that NoSQL people do not like to use the term database, but their prefer data store or something like that, because usually databases are considered by most as RDBMS: Oracle, DB2, MS SQL.

Google and Amazon are hardly typical. They require very high levels of scalability to a level that is extremely rare.

For most applications, RDBMSs can be made to scale sufficiently for the job at hand.

It will be interesting to see what Microsoft and Oracle come up with to try to keep their products competitive as different technologies emerge.

Exactly, Google and Amazon use data in a way that is not very common, they handle huge amounts of information in a heuristic way, they do not care about control, if the exact same query produces to different results, they do not care, as long as it is "more/less" relevant for the user, but inside an enteprise, things are pretty different: if you are processing the payroll of you company you want it to be perfectly predictable an consistent up to the last cent, and if you write a query asking for all the people that have earned USD 10,000 in the last 3 months you want the result to be exactly accurate not "more/less" relevant.

So one big question here is: how many companies are going to need this kind of "fuzzy logic/ heuristic" stuff to handle their data in a scalable "eventually consistent, smart but uncontrollable" way...

Agree with "not so fast". One important aspect that is easily forgotten is that "data outlives the applications". Some of the given examples of major users of non-relational DBs are also known for pionieers of the "perpetual beta". I can imagine however a world where the "core" of the enterprise data is managed as before in RDBMSs along with non-RDBMS representations of it fine-tuned for special purposes. Already today companies spend a lot of money just for keeping data in-sync between the different business domains. At every domain boundary there are already huge amounts of transformations "SQL" <-> Document (be it in EAI or SOA architecture styles).

I can imagine an enterprise with an "ACID" Core for each major business object and some "eventually consistent" technologies on the edges. An authoritative source of truth is however a must. Otherwise we will end up building home-made distributed transaction management to keep things up-to-date, not a nice perspective :-)</->

Using a NoSQL database does not mean you are using eventually consistency. In fact, Google's BigTable enforces consistency. The details are a bit complicated (I wrote a short article about it), but, in summary, the thread-offs distributed databases (whether the use SQL or not is not relevant in fact) have to make are pretty much constrained to the same laws. Not all choose the same trade-offs.

I think NoSQL is a new approach to use (which still needs a heavy cleaning of it's own scenarios), but it's definitely not the approach to override RDBMS. It's just another tool approach.

As in every IT topic: context is the key!

By the way, I think NoSQL are just getting "through of disillusionment" moment of it's own hype cycle:

www.gartner.com/technology/research/methodologi...