Abstract
This paper studies the processing time of cloud services using the retrying technique for fault tolerance. Specifically, the processing time of a cloud service is modeled and the probability distribution of the processing time is derived considering the effect imposed by retrying tasks in the case of service failures happening. As another important performance metric that can better reflect the Quality of Service (QoS) requirements of customers, the percentile of processing time is also studied. The percentile of processing time characterizes a threshold processing time so that a certain percentage of requests can be served in under the threshold value. Relationship between the task failure rate and percentile of processing time is studied through a numerical example. In addition, the percentile of processing time of cloud computing systems using the retrying technique is compared with that of systems using the check-pointing technique.