What do you do when you have more than 2.5 million users daily and have to deliver five nines uptime? If you’re Medio Systems, you roll your own cloud using Red Hat Enterprise Linux (RHEL), Hadoop, and IBM System x servers.

Medio Systems is in a bit of a niche market. The company provides a predictive analytics platform that takes a great deal of data and turns it into actionable data for network operators, handset providers, and application providers. The data is used to try to not only drive users to buy and/or consume more services, but also improve customer retention by providing information that helps providers understand customers better.
According to Medio co-founder and CTO Brian Lent, the company is not just providing reporting, but predictive analytics. What’s the difference? Lent says that there’s four stages:
- Query & Reporting: How many customers do I have?
- OLAP: Where are my customers?
- Data Mining: Which customers might buy more services?
- Predictive Analysis: What do we offer the customer right now?
In other words, instead of just getting reports about what users have done, Lent says that the goal for Medio is to help companies understand “how to make interaction with the consumer better.” That, in turn, drives a lot of data and response times are crucial.
The initial solution being offered by Medio, wasn’t cutting it. Medio was deploying co-located servers that let individual customers process data. According to a case study released today (PDF), the custom cloud using RHEL, Hadoop, and 60 IBM System x servers have a response time in the range of 75 milliseconds – down from 894 milliseconds. The number of transactions has increased from 41.5 per second to 1,200 – all while maintaining 100 percent availability.
One might wonder why Medio decided to deploy a custom cloud rather than using hosted cloud services from providers like Amazon or Rackspace. According to Rob Lilleness, Medio’s CEO, it’s all about uptime. “We do have some of it running on AWS… but certain larger enterprises have demanding SLAs. In fairness to today’s cloud market, it’s challenging to get an effective price point and get that high level of availability.” Lilleness did note that “it may be changing” and the company isn’t “wedded” to its own cloud – but right now, he says cloud providers “don’t deal with that kind of volume.”