Building an Open Source Software Company Around Cassandra, Seed-Funded by RackSpace: Jonathan Ellis and Matt Pfeil, Founders of DataStax (Part 4)

Sramana Mitra: Let’s come back to the pitch to Lightspeed based on which you raised your Series A. How did you evolve from there? How did you build the business?

Matt Pfeil: We built out an engineering team for both the core open source project as well as continued to evolve OpsCenter. For practical purposes, it felt like that was one of the goals of the money. The OpsCenter was our first company-owned product as opposed to completely open source.

It was the management software for Cassandra and we built out a team for it. We continued standard support offerings and then started to hire full-time support engineers as opposed to engineers who were doubling duty. It was really good, based on customers signing up for those offerings. We had more revenue than anticipated and ended up hiring more people than we originally planned to. Our original plan was to hire 6 people in 6 months but within a year, we had about 20 people working for the company. We also hired two sales people.

Sramana Mitra: Can you talk to me about what was happening on the customer side? Who were the customers? What kinds of deal sizes? All the dynamics from the sales and business development side.

Matt Pfeil: In 2010, most of the sales were still to technology-based companies. We did a lot of experimentation with pricing and structure of the deals to really figure out what customers wanted. The two big deals that I still remember were in the fall of 2010, maybe 4 or 5 months into the company’s life. A public company that has e-mail marketing for SOP’s called us and were evaluating us. In about 3.5 months, they deployed their first standard appointment. We also signed a couple of companies in the Fortune 50. Netflix started to look at Cassandra in August 2010, but they brought us in to just jump into the discussion of about 30 to 50 engineers just describing Cassandra. And that led to Netflix standardizing Cassandra as their database of choice for their entire infrastructure. That was exciting because Netflix has been a great partner to work with.

Sramana Mitra: So Jonathan, can you actually elaborate on why Cassandra? Why were these customers leaning towards Cassandra? What did the ecosystem look like and where was Cassandra positioned that was giving you the opportunity to work in the Cassandra ecosystem and build the company?

Jonathan Ellis: Cassandra builds applications that serve user bases of millions of users. If you look at traditional database technology like Oracle from the 1980’s, they are all designed around serving a single department or a single company. They are designed around a single database server and if you have more users, you need to buy a bigger server to run that on. The problem is that there’s a ceiling to how big a server you can buy. First, it starts getting cost inefficient but then even if you have more money, you simply can’t buy a single machine big enough to run your business on.

Early Internet pioneers like eBay ran into this problem. At first they bought bigger and bigger sub-machines to put their Oracle database on. However, by 2002, they couldn’t buy a bigger machine. So they had to start cutting up their database across different machines manually. So that became the state-of-the-art for building web applications and software or service.

The strategy was to manually distribute your database across a cluster of machines. The problem with this is that the techniques you use are not reusable. So eBay would have a one-off solution to do that for their bidding system but Amazon wouldn’t be able to reuse that effort or even a different eBay subsidiary like RedLaser wouldn’t be able to leverage that effort. They would have to start from scratch. Cassandra gives you a reusable technology.

The early adopters were mostly in the social media space. Facebook open sourced Cassandra, then Digg, Twitter, Reddit, and Mahalo were the early adopters. Then, there was a second generation of users like Netflix who weren’t really in the social media space but had a large and growing user base and needed to solve similar problems.

Sramana Mitra: And what is the approach that Cassandra uses to address the large user base? Algorithmically, in very brief, what is the approach?

Jonathan Ellis: It’s an extension of a concept called “system hashing”. What that means is that we decide what machines in every row is going to live on. When you add new machines to the cluster, then we have to reshuffle the minimum necessary data to that new machine.

That was one of the major factors for Cassandra adoption. Not only does it let you painlessly manage a cluster of machines as a single database but it also scales out. So you can add capacity to that cluster as your user base grows without down time. Cassandra is almost unique with no down time even for upgrades. We recently released Cassandra 2.0 so you could upgrade from 1.0 to 2.0 with no downtime and a lot our competitors haven’t been able to match that.

Sramana Mitra: So in the 2010 time frame, when you were starting to work with some of these customers, they were already adopting Cassandra. And then they needed management support and management tools and that’s what you were providing? Is that correct?

Jonathan Ellis: Yes. Technically what we were trying to deliver was a product business and the management tools and so forth. But, honestly, people were paying us because we were the Cassandra experts and so they wanted to have the support available if something went wrong. We released Cassandra 0.7 in October of 2010, so it was still a pretty new product at that time. People were understandably a little bit nervous about the deployment without some kind of a contract.

Sramana Mitra: They needed to have experts on hand to be able to manage that situation.

Jonathan Ellis: That’s right.

This segment is part 4 in the series : Building an Open Source Software Company Around Cassandra, Seed-Funded by RackSpace: Jonathan Ellis and Matt Pfeil, Founders of DataStax
1 2 3 4 5 6 7