Auditing cloud service level agreement on VM CPU speed

In this paper, we present a novel scheme for auditing Service Level Agreement (SLA) in a semi-trusted or untrusted cloud. A SLA is a contract formed between a cloud service provider (CSP)and a user which specifies, in measurable terms, what resources a the CSP will provide the user. CSP's being profit based companies have incentive to cheat on the SLA. By providing a user with less resources than specified in the SLA the CSP can support more users on the same hardware and increase their profits. As the monitoring and verification of the SLA is typically performed on the cloud system itself it is straightforward for the CSP to lie on reports and hide their intentional breach of the SLA. To prevent such cheating we introduce a framework which makes use of a third party auditor (TPA). In this paper we are interested in CPU cheating only. To detect CPU cheating, we develop an algorithm which makes use of a commonly used CPU intensive calculation, transpose matrix multiplication, to randomly detect cheating by a CSP. Using real experiments we show that our algorithm can detect CPU cheating quite effectively even if the extent of the cheating is fairly small.