On a framework for analyzing cloud system performance

Summary form only given. Cloud systems provide us a golden opportunity to develop special purpose systems that do one thing well; there is a performance gain in using these special purpose servers to process a request instead of using all things for all people servers. The question then becomes will the extra overhead created by the cloud in searching for the server more than offset the performance gain under all conditions. These issues are examined in this paper using a frame work which is also used to calculate bandwidth and response time. A technique based on non-dimensional parameters first introduced by Earl Buckingham in 1914 and widely used in studying problems in fluid mechanics and heat transfer, is applied to study cloud performance. Based on this method three parameters have been identified to characterize a cloud workload as delay bound, remote server bound, or middleware bound.