Economy cloud vendors feeling the heat - performance problems documented
I recently read an interesting article about how a number of cloud providers' services were studied by researchers at the University of New South Wales in Australia, and found to have significant performance problems. This isn't too surprising: virtualizing a highly variable mix of loads on shared infrastructure is a hard problem. Despite this, cloud computing is still a major growth industry, with Garner predicting that it will grow from $46Bn last year to $150Bn. While these numbers are debatable (especially since everyone has a different idea of what Cloud is and I suspect Gartner used the broadest one), they do indicate a growing market.
However within this market, there is definitely room for a variety of approaches to meet different needs for cost and performance. The largest cloud vendors, starting with Amazon, use commodity hardware and achieve surprisingly good results with it, but still limited by the realities of commodity hardware and other optimizations they have made to deliver service at the lowest possible price. As a result, customers who are attempting to use these services outside their design capabilities are in for a big surprise, as these Australian researches discovered. One discovery they made was that "response times on the services also varied by a factor of twenty depending on the time of day the services were accessed." At a recent Cloud Camp, I saw the resulting frustration, often as a result of customers having to overprovision to meet their performance goals, and as a result, being disappointed that cloud deployment costs exceeded what they expected. The researchers also brought up the lack of measurement and monitoring tools that allow customers to verify their performance in a production environment, something that we at ENKI consider to be absolutely essential to any business use of the internet.
ENKI was founded not on providing the best price, but rather the best value to its customers, which we measure as TCO (Total Cost of Operations), to which performance is a contributing factor. To this end, over the years we've added enterprise grade monitoring and have been improving the performance of our cloud to match our customers' expectations. Recently we've developed a higher-end cloud computing product that has the same performance as a purpose-built datacenter would have, called PrimaCloud . I'll be writing more about PrimaCloud in a future article.
TrackBack URI for this entry
