1 Comment

Summary:

If there was a NoSQL storm brewing earlier this decade, Hummer Winblad’s Mitchell Kertzman thinks it has all but died down. People thought NoSQL would blow up the SQL world, he said on this week’s Structure Show, but it might just be a nice complement.

Remember a few years ago when all the talk around emerging NoSQL databases was how they’d make history of relational databases any time that scale was an issue? Well, that hasn’t exactly happened yet, and Hummer Winblad Managing Director Mitchell Kertzman has a theory as to why: In most cases, NoSQL just isn’t that disruptive.

A database veteran of decades, Kertzman shared his thoughts on the space during this week’s Structure Show podcast. Here are the highlights of an entertaining and insightful interview, but you’ll definitely want to listen to the whole thing.

Download this episode

Subscribe in iTunes

The Structure Show RSS Feed

Really, everyone wants SQL

“If there were cloud-oriented SQL databases that could do the kind of elastic scalability that was needed, handle then volumes of data that were required, then there never would have been a requirement for NoSQL,” Kertzman said.

A big disclaimer is probably in order — Hummer Winblad is an investor in NewSQL startup NuoDB — but Kertzman said that doesn’t cloud his thinking about the next-generation database space. The firm actually passed up a lot of NoSQL companies over the years for a number of reasons, and one of them is a belief that companies — especially those running mission-critical or otherwise important applications — really just wanted a better relational database.

“I think of NoSQL as a new market opportunity,” he said, “and its size is as yet unproven.”

Relational databases, on the other hand, are a $35 billion market, and that’s less because people like the query language and more because they like things like ACID compliance. “If you’re Facebook … and you drop a status update or a like or something like that and it doesn’t show up — no big deal,” Kertzman said. “If you are a bank and you lose a deposit, even a small one — really big deal.”

Mitchell Kertzman

Mitchell Kertzman

That thing you do … that we can do, too

“There have been radical disruptive innovations — relational database was that to the old network databases. … And there was client-sever to host computing and now there is cloud/SaaS to client-server,” Kertzman explained. “… And then, because we all love disruptions because that’s what creates opportunity, there are what people think will be disruptive but turn out not to be.”

He puts a number of technological trends into this latter category, including the object database movement during the 1980s. “It turns out it wasn’t disruptive because the existing guys — Sybase and Oracle and everything — could kind of add that functionality or the amount that people people really needed,” Kertzman said. “People don’t want a religion; they want tools they can use. The object database guys were following the object religion, and the legacy guys were just adding object capability.”

He looks at MongoDB as a NoSQL database that can maybe win a lot of deals in companies that don’t need ACID compliance and don’t want to pay the Oracle premium. However, he cautioned after I noted the encroachment of companies such as Teradata, MemSQL and Heroku into its realm, “We like to invest in companies that are solving really hard problems with really new technology, defensible technology. Not defensible just in the IP ownership sense, but in that it’s really hard to do. … MongoDB, for all the success it’s had to date, is not particularly technologically sophisticated in solving hard technological problems.”

Timing disruption is a tough task

Sometimes, though, even when Kertzman thinks he’s spotted a seemingly golden opportunity, getting the timing right can be difficult. He noted the opportunity to get up higher in the stack in areas such as big data, where things like Hadoop are certainly for real but sometimes confusing for people to work with unless they have more intuitive analytics tools.

He thought Karmasphere, in which Hummer Winblad has made three investments in since 2010, was such an opportunity when it came to bringing a familiar business intelligence experience to Hadoop. Perhaps it still will be but, as I noted, the SQL-on-Hadoop movement has somewhat dampened the enthusiasm for tools like Karmasphere because companies can now use Hadoop with legacy BI tools and get faster performance by bypassing MapReduce.

“To some extent, I think what happened in that space was real Hadoop adoption happened slowly,” Kertzman said. “In other words, if Hadoop had really taken off … then the bet [on Karmapshere] would have gotten to market faster. Since Hadoop wasn’t being adopted as fast, then tools for Hadoop weren’t being required as fast.”

SQL on Hadoop, he added, was a reaction by Hadoop vendors to that slow adoption.

Feature image courtesy of Flickr user Michael Mandiberg.

You’re subscribed! If you like, you can update your settings

  1. I think of the space very differently.

    I talk to tech teams using MongoDB all the time, and I ask them why they use it — there are two reasons. First, some of them have very large clusters, and the horizontal scaling imperative mentioned is definitely a factor in their selection. However, there is a second reason — many users love the product even though *their entire workload fits on a single server*. So for these developers and apps (thousands I’d say, at a minimum), it wasn’t about scale.

    So why did they use it? Because to them it was simply the better and faster-to-develop tool for writing a production application that is database-backed, for their use cases. The data model and JSON fit much better with the programming languages we use today (JSON gives us a standards-based, language independent way to store object-style data). The dynamic schema nature of MongoDB (and quite a few nosql products) fits really well with the types of real world data we are working with today — which is often semistructured, polymorphic, object-structure, or evolving in structure. The “dynamic schema” notion is also super helpful when doing agile development and iteration — which we all do now.

    One-size-fits-all is over at this point, but I see the data layer of the technology stack undergoing perhaps the largest shift we’ve seen in 25 years — because of both nosql and hadoop as trends therein right now.

Comments have been disabled for this post