CTO and Founder Brian Bulkowski of Aerospike is interviewed by theCube with John Furrier, founder of SiliconANGLED and Dave Vellante, co-founder of Wikibon live at Open Oracle World 2012.
“Multi-core and distributed systems and flash is what the next revolution in databases will be”, said Brian Bulkowski, founder & CTO of Aerospike during an interview with theCube at Oracle Open World.
“What I saw were companies that were being strangled by scale…they were using sharded MySQL, which was the style of scalability at the time. They found that to be a really difficult chore. They had outages. The moments when they needed to scale most and needed to absorb the most traffic those were the moments that they were down. I said, “You know what? There’s got to be a way for the highest levels of scale, because the entire world is going to be moving to more and more scale, more and more data….”
“We founded the company four years ago as one of the fastest databases in the world. We focus on commodity hardware, and especially on flash storage. It’s a special focus of ours, realizing that companies like Oracle had great technology with rotational drive, and the world was going to switch to flash and DRAM. The time had come for a new database, so we’ve been selling our database primarily in the fastest, most scalable platforms in the world, which are mostly advertising platforms.
These are folks that are doing 10 billion to 20 billion impressions per month. We’ve had no downtime with any of our customers and we can usually solve, with flash storage, multi-terabyte systems with billions and billions of rows, often for user tracking and user information personalization of advertising and experiences. So, we’ve gone last year from a great revenue year into this year, more sales, more customers….
We’ve built our system from the ground up to be flash-oriented. We work with main memory and we also work with flash memory, and we don’t really support rotational disk at all anymore. We use it as a backup store, as a warehouse-style technology. That’s where we keep things like transaction logs in the database. But we don’t serve from it anymore, and that’s been true for years in our product. That’s why we can achieve things like – a recent blog post – 1 million transactions per second per server out of main memory DRAM. That’s for a several thousand dollar server, nothing fancy….
The main change is, when you think of users, they behave irrationally. They’re hard to predict. That’s the whole point of users. They do all kinds of things. So you can’t use caching technologies. You can’t assume that the most recent users will be the most recent users a few minutes from now, and the analytics you do, it might not be on any particular pattern. So, the typical tiered storage approaches, frankly, have a lot of trouble.
So, we said a better technology looking at flash and looking at distributed systems is one where you just have one database, no cache. So, our system is as fast as cache you put in front of a database, but you can build ten, 20, 100 terabyte systems with flash storage, a cost-effective way, and reduce the complexity of your architecture. No database caches, what a pleasant world…
“You’ve got every database in the world. You’ve got Oracle. You’ve got Sybase. You’ve got the old IMS stuff. You’ve got Vertica. You’ve got everything. Why would you come to us and start talking about a new database?” The answer was, there are random access patterns in their calculations, what’s called “What if?” scenarios, where you don’t know what’s going to happen.
You need to game out various different strategies. That cannot be done with rotational disk or cache optimization strategies. You have to have a very random, very fast solution. So, what that means in terms of the bottom line of a business is, let’s take our advertising customers, very cut-throat business. It’s all about the pennies that you shave on every transaction. With a, say, 25-server cluster, we’re able to serve, say, 25% of the advertising load of North America. That’s compelling.