Thought Leaders in Big Data: Interview with Brian Bulkowski and Srini Srinivasan, Co-founders of Aerospike (Part 4)

Sramana Mitra: Do you have to do anything with your software to optimize it for a DRAM-oriented system versus a flash-oriented system?

Srini Srinivasan: We have optimizations to use a flash system. We also have optimizations for DRAM systems, but they are simpler. We wrote our own file system to take advantage of flash storage. We use it as a raw device. Then we use our Aerospike file system implementation. This essentially gets you a 100x [faster] performance over traditional systems. We are able to take advantage of the speed at which you can read and write Flash, and we do large block writes. We are also able to read directly from Flash so you can get a main memory like performance while still using flash.

Brian Bulkowski: Another interesting customer is a behavioral web 2.0 company. We have a set of functionality around behavioral data types, where if you have a large object and you insert a lot of recent behavior, being able to read that back we have new optimizations of how we use storage. Those guys are in deployment soon. We also have a company doing real-time dashboarding entirely in-memory. They have a big Hadoop system on the back side as part of their analytics tier, but then they also want analytics and dashboarding occurring on a minute-by-minute basis. That is a new use of real-time for us as well. Advertising is on the far side of real-time – they need to be [in] milliseconds. But what we find as we move into analytics is that there are plenty of cases where people want to take their hour-long computations to take a second or a minute.

SM: And to be able to do something during the interaction.

BB: Absolutely.

SS: We are all about real-time in a sense that you are able to run queries on the database while there are rights happening. It is actually hard to do that. It is easier to just query databases. But we are able to do time series based queries, key value queries, etc. There are a whole bunch of database-specific actions that you can take.

SM: Financial services must have a lot of applications.

BB: It does. They also have a number of needs around reporting, security or other features of that sort. That is an area that we have actively investigated and there is a bunch of interesting uses there.

SM: Who do you see as competition? Is TIBCO a competitor?

BB: Not yet.

SM: You know they come from the real-time computing roots, and they are very deep into financial services.

BB: Absolutely. We have a board member who sold one of his previous companies to TIBCO in this exact area of real-time databases many years ago. TIBCO is a very interesting company. On the competitors’ side, we see a lot of “do-it-yourself” projects.

SM: So everything inside Facebook, Yahoo, Google, etc. is homegrown stuff?

BB: Yes.

SM: Have you explored going to them?

SS: We have sold to some of them, but typically to companies that are sub-partial to those companies, not the core. We enable companies to build technology using ours and grow at a fast scale. If you add up all of the actual interactions happening across our deployment, it comes very close to what Google is doing every day. That is the level at which we can do our stuff.

SM: For example, today Facebook has a bunch of content and features. Facebook’s advertising platform doesn’t give you a lot of analytics. The analytics of Facebook’s advertising is actually elementary, I would say.

BB: Facebook launched a project last year called Facebook Exchange. What they do now is they take some of their ads and they place them on some of these open markets, using AppNexus or Trade Desk. All of that load for advertising is coming out into our customers. When Facebook decided they were going to start doing this, our customers were very easily able to add capacity, in some cases double capacity, in a matter of months.

We often look at companies that are competing with the Googles and the Facebooks. Those companies have so much of their own solutions and internal engineering teams that they are very proud of. When we started working with AppNexis there were 30 people and now they are 400. They are competing with Google successfully, partially due to our software.

This segment is part 4 in the series : Thought Leaders in Big Data: Interview with Brian Bulkowski and Srini Srinivasan, Co-founders of Aerospike
1 2 3 4 5 6 7