Early Warning System For Cloud Computing

Generally speaking, when something bad starts to happen you want to know as early as possible. For example, if you catch cancer early, you can begin treatment and improve your odds of nipping it in the bud. Similarly, if you have advance notice that a tornado is coming, you have time to move to a safer place. Now researchers have developed a tool that can identify cloud computing anomalies – hopefully allowing computer hosting infrastructures to prevent such anomalies from becoming major problems.

First, a quick recap of cloud computing. Cloud computing allows lots of different users to tap into a large-scale computer hosting infrastructure. This gives them access to computing power they wouldn’t otherwise have. Okay, back to the new stuff…

Cloud computing can run into problems, ranging from bottlenecks that slow operating speed to full-bore host failures. That’s bad if your company relies on a host infrastructure that fails. It is also bad for the company that operates the host infrastructure (since they’ll have to pay for failing to provide adequate service.)

But what if you could catch these problems before they became, well, problems? It would give the hosting infrastructure a chance to diagnose what is going on and take steps to minimize (or possibly prevent) any potential damage. That’s a great idea, right? However, (surprise!) it’s complicated.

The problem is that in order to identify abnormal activity, you have to be able to define normal activity. Since host infrastructures have lots of clients doing lots of different things at lots of different times, it’s tough to accurately define what constitutes “normal” behavior.

This is where Helen Gu comes in. Gu, an assistant professor of computer science at NC State, worked with NC State student Yongmin Tan and Haixun Wang of Microsoft Research Asia to develop a collection of models that examine activity in hosting infrastructure under many different scenarios. This effectively allows the models to have multiple definitions of what constitutes normal behavior – and makes it much easier to accurately identify anomalies that could blossom into major problems.

The work will be unveiled later this month in Zurich. Hopefully it can help us avoid cloud computing traffic jams.