To be completely honest, I have never heard the term "NewSQL" before in my life. After a quick Google, it seems that the term basically means "SQL databases that go FAST LIKE NOSQL." (Though it's not really popular.) Which is, once again, completely missing the point.

I can't speak for professional devs - I'm just a hobbyist - but I'm not developing for "NoSQL" databases like CouchDB and Redis because I want to do SUPER WEB SCALE. I'm developing for CouchDB and Redis because they offer truly interesting alternative ways to look at your data. (In fact, Redis is what got me interested in data structures.) And if I wanted to write a project that used data that makes sense from a relational standpoint, SQLite (for development) and Postgres (for production) are just fine with me.

There are a lot of people who would like a full-fledged relational database that scales up to large clusters. Just because you're not one of them doesn't mean they don't exist, or that they're "completely missing the point."

I was referring more to the marketing aspects of the term (and proposing their solution as a counterpart to "NoSQL"), than the concept itself. There are a lot of people who think that "NoSQL" is all about magic scaling sauce, and this isn't really helping.

I don't really understand why one would accept the trade-offs of "NoSQL" if you aren't interested in "magic scaling sauce" (and clustered redundancy).

What's the advantage here? Less rigorous schema, moving data modeling and constraints into the more complex application code that's much more likely to have bugs?

I can't tell you how many times database constraints have caught application bugs with fatal data correctness consequences on the products we've worked on.

The only reason I can see for abandoning those declarative constraints is because of the scalability issues one must contend with (and even then, I'm somewhat unconvinced that they matter for all but the highest-load entities).

Some the the databases grouped under the over-broad NoSQL umbrella have reasons to use them other than magic scaling sauce. Redis, for example, is a ridiculously fast in-memory database that gives you data structures like sets, lists, hash tables, strings, numbers, and so on. It's tremendously useful for some problems.

For example, at Greplin we're using Redis for our document analysis pipeline. We're pushing thousands of documents through Redis every second, using it to durably hold data while it gets passed around through our various back-end worker server clusters. We've been very happy with how Redis has been performing; it handles the write-heavy workload beautifully, and is way more resistant to data loss than most systems for doing this sort of thing. We'll be even happier when it gets server-side Lua scripting in a stable release, and we can define our own commands. The point is, though, that this would be crazy to do with a traditional SQL database, or even something like Riak or CouchDB or most of the other NoSQL databases. Redis fills a different niche, and filling a lot of cool niches is what I think the NoSQL stuff should be about.

OK, you're rational. Unfortunately you're in the minority.

Exactly. I use MongoDB because schema-less databases make prototyping my applications easier. Migrations became a pain point.

For those wondering about it, don't bother searching for 'NewSQL'. It's a buzzword that doesn't actually mean anything other than 'it's not mysql, pgsql, oracle, or mssql.' They claim to improve on the standard DBs, but every single DB does it in a different way. The name is useless.

Instead, read the 'who we are' and 'how it works' bits on NuoDB's page. They were much more informative.

I think it sounds exciting and fresh... And doesn't seem to be open source at all.

I'm not allergic to paying money for software, but I've found I'm happiest when using something that I can guarantee will survive, even if may have to do so by my own hand. Commercial products have a tendency to be a pain in the rear.

Thanks for the heads up (was looking for more real info and not the damnable 'beta signup' form)

This sounds like SQL ontop of Cassandra (for a more or less mental-map if you are looking for one):

  Our distributed non-blocking atomic commit protocol allows database 
  transaction processing at any available node. The system is 
  designed to provide a consistent view of data, high availability, 
  and the ability to detect and manage partitions in a predictable   
  and consistent manner. Query processing scales with the number of 
  available nodes, storage scales according to the choice of storage 
  manager of which there are many. Storage managers are key/value 
  storage engines that also operate as a distributed in-memory cache 
  and participate in the commit protocol to ensure durability 
  guarantees.

Given that they seem to favor consistency in the CAP model, it would be closer to Hbase in this regard. Cassandra is, by design, eventually consistent and can not offer these same guarantees.

This is false. Cassandra can offer any level of consistency you desire, on a per-operation basis. See the "ConsistencyLevel" section of http://wiki.apache.org/cassandra/API. The tradeoffs of the CAP theorem, of course, apply. But if you want 100% consistency (availability and partition-tolerance be damned) it's there for you.

There is a general misconception that the an entire DBMS must pick tradeoffs based on the CAP theorem. (IE, you can use a database that offers availability, but can NEVER offer immediate consistency) Instead, a DBMS may offer different CAP guarantees per operation.

I don't think so. Does Cassandra provide guarantees regarding the ordering of queries? Cause I couldn't see that.

Another problem I can see is when you write to 3 nodes, have W=2 and a write succeeds only one one node. Then later on through read repair the value propagates to the other 2 nodes. So our write "failed" but really in the end succeeded. I'm not too much into the Cassandra details but those two things immediately came to my mind when reading some of the wiki pages.

It boils down to not using/offering algorithms like 2-Phase Commit or Paxos but more sloppy quorums.

So from what I can tell, Cassandra can not guarantee strong consistency. But then again I could be wrong and missed something.

Taken directly from the page I linked from above:

"Note that if W + R > ReplicationFactor, where W is the number of nodes to block for on write, and R the number to block for on reads, you will have strongly consistent behavior; that is, readers will always see the most recent write."

Could you explain why this is incorrect? It seems the Cassandra devs use the definition of "consistency" from Brewer's paper, and Cassandra offers this level of consistency.

Your point about read-repair is correct when using QUORUM consistency level, but even then you can avoid that by using the ALL consistency level.

QUORUM reads & writes in Cassandra are still stronger than guarantees given by async master/slave systems like MySQL, MongoDB, HBase, or Redis.

Thanks for the correction -- am still fuzzy on the details of Cassandra compared to HBase.

All the NewSQLs I've seen are shared-nothing cluster databases; that's basically the definition of the term. (Of course, that technology was invented in the 1980s and then forgotten, so it's new to somebody.)

As for open source, something you can't say is that open source communities are unable to develop this kind of software.

Given that the CEO worked with the author of the 1986 paper about shared-nothing clusters, you're probably right.

"NewSQL" basically means "old guy desperately trying to stay relevant in a post-relational world" See also: VoltDB.

Now with Cloudbursting!

Looks like they are doing a pure enterprise push: NuoDB, Inc., the thought leader in cloud database technology, ...

I'm a big fan of relational theory and I'm always happy to see any activity in this space. However, their "revolutionary" distributed model seems to be similar to something which is already out there, open source, and ready for use:

http://bandilab.org/blog/2011-06-22.Ostap.Running_on_Multipl...

or maybe I just don't have enough information about the nuodb ?

Nope. Not enough info on NuoDB.

NuoDB doesn't have a centralized transaction controller. It doesn't have a centralized anything. And it isn't "like" anything else -- all new. It's a distributed database without distributed transactions.

In a nutshell, it isn't based on pages or disks but distributed objects. Each distributed object, called an atom, can exist in as many places as necessary. Each atom instance knows of the other instances and replicates changes. Transaction nodes do SQL and are diskless. Archive nodes listen to replication and store serialized atoms on disk, S3, or some other KV store. If a transaction node needs an atom, it gets it from another transaction node or, at last resort, from an archive node that fetches the atom from disk, S3, or wherever.

fair point about the transaction controller, this is indeed something which is not yet distributed in bandicoot. Though, I don't think it's a big deal as the tx is really simple and lightweight.

Anyway, let's go back to nuodb :) I'm really interested on how it manages the whole ACID. I did not get it from the page nor from your reply.

For simplicity, let's pick on durability only. The archive nodes listen to replication and store serialized atoms on disk. At the same time the atoms replicate the changes to other instances. Is the durability achieved by a synchronous replication?

Thank you for the clear and concise description. Your website really needs to say that, plus have a diagram.

Are there NuoDB-specific drivers/connectors for languages? What interface would I use in Python or Ruby? What would I need to install?

Has anyone actually tried this wonder-machine or is it Vaporware?

I remember submitting an evaluation request almost a year ago but never received a response.

Their homepage and blog don't seem to have changed since then either; no real information anywhere, only truck-loads of PR fluff.

Their claims reminded me of Jim Starkey's history with the CAP theorem: http://pl.atyp.us/wordpress/index.php/2010/03/cap-and-leases...

in case anyone else was wondering: the key CAP-related "compromise", from "how it works", is that on partition it supports transactions on only one sub-set of nodes (the "healthiest"); the others provide read-only snapshots.

http://nuodb.com/how_it_works.html (see "Availability and Resiliency").

In other words, it's plain old CA.

Anybody know why they changed the name? Nimbus seems more pronounceable.

To add validity to their new term "NewSQL" possibly? (I don't know, just guessing).

To the "NewSQL" term, I don't really have a problem with it... from their "how it works" it sounds like they are genuinely doing something interesting and cool with SQL. I have no qualms with them calling it "NewSQL".

I don't know why marketing terms get people so mad, sometimes they make sense.

I'd also point out that the first few times people used "NoSQL", I imagine there were comments along these same lines talking about how silly it was.

It isn't like they named it "WebScaleSQL" :)

Marketing statements notwithstanding, the name change was the result of something really boring -- a name conflict with another company.

Oh, "New-oh dee bee".

Nimbus Data, a maker of of flash storage, wasn't too happy about NimbusDB and the brand confusion

can't wait to read the white paper

I'm guessing you're referring to their claim of being able to do distributed transactions?

Data is spread out across transaction and storage nodes and it always available. When network partitions occur NuoDB continues offering services via the most available and healthy segment while the other segments provide a consistent read-only snapshot while awaiting reconnection at which point they will synchronize with their counterparts. Clients can simply reconnect to the active partition of the cluster to continue processing updates.

I'm not sure how they can ensure that network partitions only occur between fully replicated sets (i.e. how can all the data be available in all locations?)

Alas, I can't downvote this link.

Can anybody explain why I should go with this instead of Pg?

To the submitter or anybody else sufficiently knowledgeable: When will ORMs get support for this? I use SQLA but could port Django or ActiveRecord code.

disclaimer: I have a bias as we could be seen as a "competitor".

If I understand correctly, their SQL engine can scale through the use of a proprietary peer to peer technology. This would be new and would indeed make it possible to solve bottlenecks and scale the SQL database to a very large number of nodes.

However, I think it misses the point where SQL is simply not always needed. They will always be slower than a NoSQL engine that scales well.

Your example of ORMs is on spot: do you really need SQL as the back end of an ORM? If not, why pay the relational tax?

No silver bullet.

The question really isn't SQL but ACID transactions. If you can do ACID transaction, SQL drastically reduces the amount of data transmitted.

There is no intrinsic reason that ACID transactions, with or without SQL, can't scale, just that up until now, it hasn't.

And there's a reason that until now, it hasn't. From the beginning of time, academic computer "scientists" have confused the terms serializability and consistency. In short, serializability is a sufficient condition for consistency, but it isn't a necessary condition. If you design a system that enforces consistency without requiring serializability, it scales. Period. Legacy RDMSes don't work that way, but that's their problem.

Huh. Your insight is an interesting one, but I am having trouble understanding how it is actually useful.

If I may reduce to a concrete example:

Posit shared variables x and y, both zero, and simultaneously do (on threads T1 and T2)

    T1: atomic { y = 1; return x; }
    T2: atomic { x = 1; return y; }

where 'atomic' indicates a transactional operation.

Traditional ACID rules would state that either T1 takes effect before T2 (and so (T1, T2) return (0,1)) or T2 before T1 (so they return (1,0)). Other interleavings are not permitted -- returning (0,0) or (1,1) would violate consistency.

What it seems to me that you are saying is that both (0,0) and (1,1) might be OK, depending on your application domain -- maybe the app can specify a consistency rule that says that (1,1) isn't OK. Then you can design databases that follow application-directed consistency rules.

At first blush this seems impractical, so I am hoping you can elaborate on the point you were trying to make.

The C in ACID is consistent, not serializable (i.e. it's not ASID). Consistency means that declared consistency constraints are enforce -- no dirty writes, no violations of unique indexes, referential integrity, and anything else that the bells and whistles supports.

Let me give a simpler example: Database with one table of one field, a number. One transaction: count the number of records and store that number.

A serializable system will force one zero, one one, one two, etc. But a consistent system can have two zeros and no ones. Why? Because that's what each concurrent transaction saw? Nothing wrong with that. But if application semantics dictate that each value must be distinct, then put a unique index on the number and the system will enforce uniqueness. Automatically enforcing "auto-magic" constraints that nobody cares about is why serializability destroys scalability of distributed system.

In NuoDB all messaging is asynchronous and batched, making it very fast and efficient.

For more than you want to know, see http://www.gbcacm.org/sites/www.gbcacm.org/files/slides/Spec...

Someplace there's even an audio recording which I recommend if you're into self-abuse.

I'm lost here as well. Let's imagine I have COUNTER table of one numeric field VAL. All I do in transactions is:

update COUNTER set VAL = VAL + 1

Let's assume the initial state is one entry with VAL = 0. If I run the update statement in two concurrent transactions I'd expect the result to be 2 when both transactions complete. However, I don't understand how that can be achieved without serializing those transactions.

The system recognizes that two concurrent transactions are trying to update the same record. If the first commits, the second will return an update conflict. The application can then restart the transaction.

These are the same MVCC semantics I used in Rdb/ELN (1984), Interbase (1986), Firebird (1999), and MySQL Falcon (2006). The implementation, however, is wildly different.

thank you for the explanation. I'm really sorry but I don't agree that this is the way going forward. You claim that serializability is not needed to achieve consistency. I agree up to the point that it's only possible if the conflicts are recognized and transactions aborted.

What is the benefit here? As an application developer I need to restart the transaction and hope it's not going to fail again - hence I'm serializing the execution on my own...

On top of all this, when it comes to constraints in relational model they can be complex and the probability of failing transactions is just going to grow. I can see this working only if I start relaxing on my constraints and redirecting transactions in such a way that conflicting transactions are coming to the database already serialized.

What about restarting transactions within NuoDB transparently as soon as an update conflict has occurred?

We'd be happy to address these issues if you are interested.

It's not clear where we're misunderstanding each other, but for clarity NuoDB manages such things as atomic operations and update conflicts without reference to the application. It's an ACID database.

Give us a call or attend a webinar (eg tomorrow) if you want to take a deeper look at it.

Barry Morris, NuoDB

So running with your example, assume I have a transaction that adds 5 to a column value and then reads the value back.

If I start with a value of 0, then run my transaction twice, can both transactions return 5? Or is it guaranteed that one will return 10?

[sorry. replied earlier incorrectly]. one of the two transactions would abort in this case. snapshot isolation needs to check that any data mutated in the transaction were not also mutated externally. in the example you are replying to, there is no mutation, and so no problem, but in your example there is. see http://en.wikipedia.org/wiki/Snapshot_isolation

note that it's only mutated values that are checked for conflicts, and only against other mutations (this is why it is efficient - the number of checks required is small). so you can get weird behaviour when multiple values are read while different transactions change each - there's a good example in the link above. this is called "write skew".

the whole approach is, in a sense, exploiting poor phrasing of the ansi sql-92 standard, which doesn't actually require serialisation even though that is the most natural way to interpret it (as far as i understand things). so you can think of MVCC as "exploiting a loophole" that leads to a more efficient system, but one that is less intuitive. on the other hand, this is not new - it's already the standard behaviour for postgres, oracle, sql server, etc.

Thanks for the clarification. However, this does not sound like something I could use in practice.

If I have two transactions and one of them is aborted because of the changes done by the other transaction what am I as a developer supposed to do? Retry? I hope not, because a retry is in other words serializing the execution. One after another.

So, if I have a system with a lot of concurrency (i.e. bank accounts and transfers) I'd better not use NuoDB because I'd get a lot of transfers aborted, not good. If I have a system with a very little write conflicts I'd go for serializability because that gives 100% consistency and will be fast anyway due to very little conflicts.

From the wikipedia link you posted, it is fairly clear that the Snapshot isolation is good when you don't need consistency. For that you'd have to either abort every time there is a conflict or introduce write-write conflict (ie. serializing).

(1) serializability will not be "fast anyway". it will be slow. this is the problem with, for example, mongodb's global write lock.

(2) you would not get "lots" of aborts with bank accounts because each transaction is, typically, to a different account. you're only going to get a problem when two processes try to change the same person's account at the same time.

(3) what this provides is standard-compliant ACID sql, the same as postgres, oracle, sql server, etc. if you use any of those and don't have retry in your code for when transactions fail then you're already in a mess.

i am not associated with this project, but what you're saying doesn't really make sense. as far as i can see, you're criticising it for being the same as everyone else in the standards compliant, sql world.

I'm not criticising it for being the same as everyone else. I'm just not sure how the whole thing works. I don't see enough information provided on the site, yet a lot of statements about providing full ACID and scalability.

re 1) and 2). I think they are related. The serialization can be done on account level. There is no need to have a single global write lock. This also implies that if the bank account transfers are typicaly to different accounts the serialization will be as often as the aborts. Hence very rare and the system will perform equally good in both cases. However, the serialization does not require further retries and doesn't force the application programmers to workaround the problems.

My understanding of the MVCC at work is that writes must effectively be serialized, but that pure-read transactions never have to block, as they can view a consistent snapshot of the state at any time.

This would imply read-scaling, but I don't see how this makes writes go any faster without reducing consistency. What am I missing?

> Your example of ORMs is on spot: do you really need SQL as the back end of an ORM? If not, why pay the relational tax?

That would be because "ORM" stands for "Object Relational Mapper".

What I was implying is that the database back-end does not necessarily have to be relational for this to be done.

The fun part of ORMs isn't that the relational algebra of SQL is mapped into your language, but that the relations between objects in your object-based language are mapped onto SQL. See Axiom for a great example of what this can mean.

At a glance, they say that they support most of SQL99, so semantically it wouldn't be hard to adapt ActiveRecord or Django to work with it, assuming the drivers were easy enough to use.

I guess they billed it as "NewSQL" not "NoSQL" which is pretty absurd.

I'd vote that you go with Pg, because it's Awesome MostAwesomeDude, at least until this has been around for a while.

If you don't know that you need something like this, then you probably should be using Postgres or something instead. It's mature and usually plenty fast enough. (Unless it isn't, in which case you can feel free to start considering fancy databases like this one.)