PEER-TO-PEER CONVOLUTED FAULT RECOGNITION TO CONQUER SINGLE-POINT STOPPAGE IN CLOUD SYSTEMS

Abstract

Cloud computing provides solutions with unlimited capabilities in such a way to meet the demand of rapidly growing consumerism. Therefore it receives more and more attention from huge players such as backing sectors, educational institutions, government divisions, top-notch enterprises, research organizations etc. At the same time, the unprecedented transformational nature of the Cloud makes it vulnerable to various unpredictable and significant security violations and Byzantine risks. Moreover initial setbacks in detecting byzantine errors often allow it to propagate considerably. Since the byzantine error can simultaneously propagate through various paths it is not feasible to detect all the propagation paths. Many existing solutions fail to detect byzantine faults since they consider monitoring for reactive symptoms rather than capacitating a proactive detection. We consider the security concerns in Cloud as Pure Byzantine problem since it often aims to compromise hypervisors in Cloud initially to act as a decoy to attack the Cloud. Since hypervisors in Cloud exhibit ever growing number of vulnerabilities, they are prone to devastating attacks such as EDoS. Unlike a VM failure, if a hypervisor in Cloud is compromised it allows the error to be propagated throughout the Cloud. Hence in this paper a peer-to-peer hypervisor based verification and validation has been proposed. It performs smart monitoring in the entire set of Hypervisors through promoting intercommunication among them to detect single point failure. The communication involves sending and receiving integrated validation and verification challenge among the Hypervisors at regular interval to improve the possibility of detecting both Byzantine and Pure Byzantine faults. Validation involves checking whether the data generated by the Hypervisors are genuine or not using hash comparison. Whereas Verification involves, checking whether the Hypervisors are performing to the set benchmark or not through performing a bound check for strategically chosen SLA metric.