The wonders of virtualization are touted daily, and with good reason. But it isn’t always the best solution for all workloads. Or all companies.
Often startups short of money turn to service providers that sell infrastructure as a service so they can spend money on software development rather than build a data centre.
But as a predictive analytics company called Sociocast found out, not only can a young company outgrow its office, it can also outgrow virtualization.
The company was saved by a Vancouver-based provider with a back-to-the-future solution: Running applications over bare-metal servers, but still in the cloud
As Sociocast CEO Albert Azout tells it, his company was started in 2010 with the idea of analyzing audience data from media companies that could be bought. It changed to offering predictive analytics of so-called event data – events being, for example, clickstream data, Web page views or mobile device data.
The massive data would be sent from advertising and media companies, social media and gaming Web sites, retailers or governments, and processed using Hadoop and other database processes on a cloud provider’s infrastructure running on 40 or so virtual servers. Sociopath crunches the number and comes up with recommendations for anything from detecting suspicious online activity to predicting retail customer buying patterns.
Pricing is based on a bundle of services: It starts at US$39 a month for analyzing up to 1 million events a month, handling up to 10,000 API calls to Sociocast servers for predictions and classifying up to 25,000 unique URLs. Other packages run at $129, $199, $749 a month.
Sociocast managed to sign up 20 customers.
However, the skies over this nirvana began to get, shall we say, cloudy over a year ago.
“We process huge amounts of data on a monthly basis, upwards of three hundred billion transactions on behalf of our clients, so every microsecond counts,” Azout said. But latency began to be a problem: It would take a longer time to run some calculations, which wasn’t good for business. However, Sociocast was using as much CPU capacity as it could buy.
“From a cost perspective it wasn’t making any sense,” said Azout. “And from an infrastructure perspective we weren’t fully maximizing throughput of the system.”