On Versant`s technology. Interview with Vishal Bagga.

“We believe that data only becomes useful once it becomes structured.” — Vishal Bagga

There is a lot of discussion on NoSQL databases nowdays. But what about object databases?
I asked a few questions to Vishal Bagga, Senior Product Manager at Versant.

RVZ

Q1. How has Versant’s technology evolved over the past three years?

Vishal Bagga: Versant is a customer driven company. We work closely with our customers trying to understand how we can evolve our technology to meet their challenges – whether it’s regarding complexity, data size or demanding workloads.

In the last 3 years we have seen 2 very clear trends from our interaction with our new and existing customers – growing data sizes and increasingly parallel workloads. This is very much in-line with what the general database market is seeing. In addition there was request for simplified database management and monitoring.

Our state of the art Versant Object Database 8 released last year was designed for exactly these scenarios. We have added increased scalability and performance on multi-core architectures, faster and better defragmentation tools, Eclipse based management and monitoring tools to name a few. We are also re-architecting our database server technology to automatically scale when possible without manual DBA intervention and allow online tuning (reconfigure the database instance online without impacting applications).

Q2. On December 1, 2008 Versant acquired the assets of the database software business of Servo Software, Inc. (formerly db4objects, Inc.). What happened to db4objects since then? How does db4objects fit into Versant technology strategy?

Vishal Bagga: The db4o community is doing well and is an integral part of Versant. In fact, when we first acquired db4o at the end of 2008, there were just short of 50,000 registered members.
Today, the db4o community boasts nearly 110,000 members having more than doubled in size in the last 2+ years.
In addition, db4o has had 2 major releases with some significant advances in enterprise type features allowing things like online defragmentation support. In our latest major release, we announced a new data replication capability between db4o and the large scale enterprise class Versant database.
Versant sees a great need in the mobile markets for technology like db4o which can play well in the lightweight handheld, mobile computing and machine-to-machine space while leveraging big data aggregation servers like Versant which can handle the huge number of events coming off of these intelligent edge devices.
In the coming year, even greater synergies are being developed and our communities are merging into one single group dedicated to next generation NoSQL 2.0 technology development.

Q3. Versant database and NoSQL databases: what are the similarities and what are the differences?

Vishal Bagga: The Not Only SQL databases are essentially systems that have evolved out of a certain business need – The need was essentially to have horizontally scalable systems running on commodity hardware with a simple “soft-schema” model for example social networking, offline data crunching, distributed logging system, event processing systems etc.

Relational databases were considered to be too slow, expensive and difficult to manage and administrate, expensive and difficult to adapt to quick changing models.

If I look at similarities between Versant and NoSQL, I would say that:

Both systems have designed around the inefficiency of JOINs. This is the biggest problem with relational databases. If you think about it, in most operational systems relations don’t change e.g. Blog:Article, Order:OrderItem, so why recalculate those relations each time they are accessed using a methodology which gets slower and slower as the amount of data gets larger. JOINs have a use case, but for some 20%, not 100% of the use cases.

Both systems leverage an architectural shift to a “soft-schema” which allows scale-out capability – the ability to partition information across many physical nodes and treat those nodes as 1 ubiquitous database.

When it comes to differences:

The biggest in my opinion is the complexity of the data. Versant allows to you to model very complicated data models seamlessly with ease whereas doing so with a NoSQL solution would be much more effort and you would need to write a lot of code in the application to represent the data model.
In this respect, Versant prefers to use the term “soft-schema” –vs- the term “schemaless”, terms which are often interchanged in discussion.
We believe that data only becomes useful once it becomes structured, in fact that is the whole point of technologies like Hadoop, to churn unstructured data looking for a way to structure it into something useful.
NoSQL technologies that bill themselves as “schema-less” are in denial of the fact that they are leaving the application developer the burden of defining the structure and mapping the data into that structure in the language of the application space. In many ways, it is the mapping problem all over again. Plus, that kind of data management is very hard to change it over time, leading to a brittle solution difficult to optimize for more than 1 use case. The use of “soft-schema” lends itself to a more enterprise manageable and extensible system where the database still retains important elements of structure, while still being able to store and manipulate unstructured types.

Another is the difference in the consistency model. Versant is ACID centric and Versant’s customers depend on this for their mission critical systems – it would be nearly impossible for these systems to use NoSQL given the relaxed constraints. Versant can do a CAP mode, but that is not our only mode of operation. You use it where it is really needed; you are not forced into using it unilaterally.

NoSQL systems make you store your data in a way that you can lookup efficiently by a key. But what if want to lookup something differently; it is likely to be terribly inefficient. This may be okay for the design but a lot of people do not realize that this is a big change in mindset. Versant offers a more balanced approach where you can navigate between related objects using references; you can for example define a root object and then navigate your tree from that object. At the same time you can run ad-hoc queries whenever you want to.

Q4. Big Data: Can Versant database be useful when dealing with petabytes of user data? How?

Vishal Bagga: I don’t see why not. Versant was always designed to work on a network of databases from the very start. Dealing with a Petabyte is really about designing a system with the right architecture. Versant has that architecture just as intact as anyone in the database space saying they can handle a Petabyte. Make no mistake, no matter how you do it, it is a non-trivial task. Today, our largest customer databases are in the 100’s of terrabyte range, so getting to a Petabyte is really a matter of needing that much data.

Q5. Hadoop is designed to process large batches of data quickly. Do you plan to use Hadoop and leverage components of the Hadoop ecosystem like HBase, Pig, and Hive?

Vishal Bagga: Yes, and some of our customers already do that today. A question for you: “Why are those layers in existence?” I would say the answer is that most of these early NoSQL 1.0 technologies do not handle real world complexity in information models. So, these layers are built to try and compensate for that fact. That is the exact point where Versant’s NoSQL 2.0 technology fits into the picture, we help people deal with complexity of information models, something that 1st generation NoSQL has not managed to accomplish.

Q6. Do you think that projects such as JSON (JavaScript Object Notation) and MessagePack (binary-based efficient object serialization library ) play a role in the odbms market?

Vishal Bagga: Absolutely. We believe in open standards. Fortunately, you can store any type in an ODBMS. These specific libraries are particularly important for current most popular client frameworks like Ajax. Finding ways to deliver a soft-schema into a client friendly format is essential to help ease the development burden.

Q7. Looking at three elements: Data, Platform, Analysis, where is Versant heading up?

Vishal Bagga: It is a difficult question as database and data management is increasingly a cross cutting concern. It used to be perfectly fine to keep your Analysis as part of your off-line OLAP systems, but these days there is an increasing push to get Analytics to the real time business.
So, you play with Data, you play with Analytics whether you do it directly or in concert with other technologies through partnership. Certainly, as Versant embraces Platform as a Service, we will do so through eco system partners who are paving the way with new development and deployment methodologies.

Related Posts

– Objects in Space: “Herschel” the largest telescope ever flown. (March 18, 2011)

– Benchmarking ORM tools and Object Databases. (March 14, 2011)

– Robert Greene on “New and Old Data stores” . (December 2, 2010)

– Object Database Technologies and Data Management in the Cloud. (September 27, 2010)

##