Is “In-Chip” Data Processing the Next Revolution?
While in-memory computing has been making the hype circuits recently, analytic startup SiSense says their technology takes data processing to the next level: in-chip.
The company launched their latest Prism 10x product – an analytics package that they say provides non-technical users with the ability to analyze 100x more data at ten times more speed than current in-memory offerings. The reason they can make these claims, they say, is due in their in-chip data processing technology.
“The difference here is in the latency cycles,” Bruno Aziza, former Worldwide Managing Director for Analytics, Cloud & Big Data at Microsoft and Current VP of Marketing at SiSense told us. “When you’re throwing a query to disk, it takes a million cycles. RAM is about 450 cycles. And you advance now to the core of the machine and you see that the L1 cache is the fastest way to process data – it processes it in three cycles.”
The SiSense query kernel is built to run in the L1 cache, says Aziza, noting that their Prism 10X caching algorithms decide in real-time how to use machine capacity to store, compress, and access the data – all done on commodity hardware. This is significant because executives at SiSense believe that there is a big gap in the market where commodity hardware plays that is currently being underserved.
“Right now we see that there is a huge void.” SiSense CEO, Amit Bendov told Datanami. “There are very high-end solutions that work well but are beyond the means of most companies (and even the larger companies are not eager to spend money these days), and at the lower end, the simpler solutions that are agile, but can’t scale and provide those [high-end] capabilities. But 90% of the companies are in between – they like the agility, but they also want the scalability. That’s what we’re aiming for – to be user friendly, but also solve non-trivial problems where data size and complexity are concerned.”
SiSense says they accomplish these performance claims by utilizing what they call their “ElastiCube” technology, which is essentially a columnar database management system, where each field is individually stored in a memory-mapped file. For executed queries, only the fields referenced in any given query are loaded (as directed by their algorithm), which they say leaves enough space for processing the query entirely without needing to read or write to the hard drive. Once the field is no longer needed, it is removed and the consumed space is released.
They say their ElastiCube approach provides several advantages outside of fast query response time under modest hardware configurations. Because the ElastiCube does not require pre-aggregation for fast query response, they say that the actual creation of an ElastiCube takes a fraction of the time to set up compared to a data mart or an OLAP cube. Additionally, as is the case with columnar databases where columnar data is of uniform type, the data lends itself to high levels of compression, which SiSense says means less hardware is needed than a traditional BI data base.
SiSense boasts over 400 customers, including healthcare giant, Merck, who they say uses the SiSense software in the cloud for processing vaccination data. They say that retailer, Target, is using their offering for such things as theft and fraud prevention, with analysis being done on where theft is most likely to happen using data from its multiple locations. Other customers include Caterpillar, Philips, Yahoo!, and Equipment Data Associates.
In April, the company took in a $10 million dollar funding round led by Battery Ventures, following a period of impressive growth, where the company says it grew 300% two years ago, and 520% in 2012.
“Our biggest challenge right now is that we can’t hire quickly enough,” says Bendov, who says that they would like to invest more in sales, marketing and support. Additionally, he says there is plenty of room for their offering to grow. “We still have substantial, dramatic ideas on where we want to take the product – it’s very innovative as it is right now, but there is still more coming up.”
Aziza says that SiSense sees a perfect storm in the market which their technology is well positioned to leverage. “Storage is very cheap, CPU is very cheap. Network and bandwidth are pervasive, yet when you look at the cost of analyzing data, it’s still exceptionally expensive,” he explains. “I call it ‘analysis paralysis,’ where the cost of analysis has been paralyzed. Companies are still paying the same price that they were 30 years ago – where everything has come down, the cost of analysis is still crazily expensive.”
Ultimately, says Aziza, they believe that in-chip analytics will fuel a disruption in the market that will change how people think about data. “The answer is in CPU and nobody is doing it.”