May 1st, 2013 NYC MySQL Group

On Wednesday, May 1, 2013, OLC attended the New York City MySQL group’s event, Future of 21st Century Databases: CEO/CTO Discussion Among Database Superstars. The event featured a superstar panel of Eliot Horowitz, CTO and co-founder of 10gen and MongoDB, Barry Morris, founder and CEO of NuoDB and Bob Widerhold, President and CEO of Couchbase. Eric David Benhari of AppNexus moderated the event.


The event started with the panelists quickly introducing themselves and their company—almost like an elevator pitch. “We have the least well-known name,” Barry Morris of NuoDB began. “We just recently got on the market. We look like NoSQL, but we’re all SQL inside.”

“Couchbase is regarded as leaders in NoSQL,” Bob Widerhold of Couchbase said. “We are leaders in key value database and now, we’re direct competitors with 10gen and MongoDB. NoSQL has grown very quickly and the type of apps created today is very fast.”

“At MongoDB, we asked ourselves about what we’re struggling with and we came up with scaling and relatability. When it came down to it, it was the inflexibility with the schema. The document model worked very easy. With scalability, it gets easier than the relational model,” Eliot Horowitz of 10gen said.

Widerhold talked about Memcached and how people were using it together with MySQL. “The first opportunity we saw was layering and we focused on very reliable scalability from a low latency perspective. It was always our strategy to involve ourselves with database. We extended our database with our merger with CouchDB,” he said.

“So Jim Starkey worked at MySQL and there was this plan to replace it,” Barry Morris said. “Jim thought, ‘We’re solving the wrong problem.’ The problem with 1980s database was that it was the transactional model was always the same—a file system. He realized that we have to get away from that. He built a p2p file system that provides transactional persistence. The system is emergence—intelligent behavior in groups, but no one in charge, you know? Our design was that it’s layered on top of key value sources. The storage architecture is quite separated from the database system,” he said.

Eric Benhari asked if there any functionality that the panelists’ databases could not replicate. “I think we certainly believe that relational databases won’t go away. We think it does things very well. If you look at changes today, MySQL databases are better aligned with scalable requirements that people are looking for,” Widerhold said.

“One is transactions,” Horowitz said. “MongoDB doesn’t have it today, but we’re planning to add that. If you can limit your transaction to a slice, sure, you can do multi- document transactions. We want to make sure scalability and ease of use is our priority.”

“It’s a SQL database. It supports most of the language you’re on. Just because a particular database has particular limitations, it doesn’t make them not scalable. We ran millions of transactions and we don’t know the limit. With respect, some of the suggestions about limitations are outdated. On transactions themselves, I think people have misunderstood them. The point of transactions is like anything else in computing—reduce cost and improve reliability,” Morris said.

Benhari read a question from the audience. “What’s the largest deployment of your database?” he read.

“We’re an operational database,” Widerhold said. “I think our largest is 80 nodes.

OMGPOP was using 120 nodes and storing hundreds of terabytes. That’s probably the largest individual clusters we have without application sharding.”

“The system does scale up to 100 nodes without sharding or manual partitioning. IN terms of database set sizes, we’re talking terabytes,” Morris said.

“The largest we know about is a government cluster at about 3,200,000 transactions per second,” Horowitz said.

“So what’s the ideal fit for your database?” Benhari asked.

“In our case, whatever you’re using, you can use us for. We’re good at low latency high concurrency. That’s what our system is good at,” Morris answered.

“We have over 400 customers now—social, mobile, advertisement, sales force, ecommerce, enterprise...we’re big in financial companies. Regarding case studies, it’s oftentimes that we’re first deployed. We provide a great solution to Memcached. We’re big in global identification stores,” Widerhold said.

An audience question was read by Benhari. “Can you discuss replacing Oracle at Orbit?” it read, directing the question to Widerhold.

“They wanted higher performance,” he answered. “They had a lot of downtime and our system is always on. The soundbyte is always on. My understanding now is that they use us for all inquiries and Orbitz is one of many that use our system.”

“Why do people migrate? One is that data models cause trouble. One thing you see commonly is complex structures. In a document database, that’s much easier to map out,” Horowitz added.

To stir up some controversy, Benhari asked the panelists what feature they envy most from each other, to the audience’s delight.

“To us, it would be transactional performance,” Horowitz answered.

“Broad, deep-seated features for developers,” Widerhold said. “We’re focused on scalability and performance and that’s what we’re good at. Regarding envy, I don’t really have anything—I mean, we’re all in different markets. I don’t see us competing.”

“I will say this,” Morris countered, “what I’ve seen happening is more convergence than divergence. “You’re seeing a convergence of these models. On the other hand, you’re seeing more and more SQL on non-SQL systems. One of the things we’d love to have is document models that 10gen employs. The other thin is that we regard our key value technology as storage technology, no as database technology. What I’d love to have is the level of storage that Couchbase has,” he said.

“How do you replicate ACID?” an audience question read.

“We think that we can go fully ACID using relational keys. We keep data on some shards and you can really do it—I mean, there are a lot of ways to do it,” Horowitz said.

“In terms of consistency, you can read your rights and we replicate that data on other nodes. If you want to perform certain actions, we have parameters to fit that,” Widerhold said.

“The issue on transactions is that the whole idea is levels of abstractions. If it can be done under the idea of abstraction, you want to do it that way because you have no choice. I believe that we’re all moving in the same direction,” Morris answered.

Benhari asked if there were any key requirements to compete in database space.

“It’s a gigantic landscape. We’re moving to next-gen Internet and there’s lots of space to make lots of people successful. Performance, scalability and reliability are somethings that I prioritize,” Morris said.

“NoSQL, in terms of database is very young. You’re starting to see a few players pull out from the pack. I think the NoSQL industry is going to be very large. It’s not going to be a single company that will dominate. There will be niche specialists. Another thing is that we’re close to a NoSQL explosion in the broad market,” Widerhold said.

“There’s going to be a lot of consolidation in the next few years. Scalability and reliability will be paramount, but developer attitudes will be equally as important. Language will also change,” Horowitz predicted.

To close out the panel, Benhari asked the panelists where they thought the future of databases were headed. Horowitz believed that it would all become one database and become a “FrankenSQL.” For Morris, it was escaping the traditional idea that schema is how one stores data. Benhari asked if big data was in any way related to NoSQL, to which the panel vehemently disagreed with. “I mean that’s such an amorphous term,” Widerhold said. “Most people think about analytics when they’re talking about big data.”