Cloud computing buzz is on the raise. Cost reduction is one of the perceived benefits of this capability. But is it really that cheap?
Yes, it's a great solution for small companies with Web-based operations, which don't have to plan for peak demand now - see "Animoto's Facebook scale-up" as an example.
We've looked into applicability of such Infrastructure-as-a-Service solution for Intel Silicon design environment, and I may blog in the future on some of the findings from this study.
At Intel, we utilize internal distributed grid of over 60,000 compute servers to run various jobs at over 80% utilization. The idea was to consider an external IaaS capacity as an extension to accommodate possible peak demands.
We've built a proof-of-concept environment using one of the major IaaS clouds. The applications we are interested in are mostly single-threaded, and are CPU bounded. We found that per-core performance of the external cloud instances is significantly lower compared to our internal compute servers. As a result, we'd need 3x-4x more instances in the cloud as opposed to the internal deployment to achieve the same throughput. The TCO for such use (calculated based on July'2009 pricing) would be significantly higher in the cloud, this is one of the reasons for us stepping back at this point in time.
I'll talk more about this study on the Cloud Computing Summit in Israel next week.
Cloud vendors - bring the performance up and the cost down, and we may come back!
Are you already leveraging external cloud capabilities to complement your computing infrastructure? What is your experience?
Till the next post,