Minimization of Expected User Losses Considering Co-resident Attacks in Cloud System with Task Replication and Cancellation

Abstract

Advances in cyber-physical systems have engendered strong needs to use cloud computing for data storage and task processing. This paper models a cloud system implementing the task replication with cancellation (TRC) technique to improve the successful completion probability of a real-time task while mitigating additional system loads and user expenses. Particularly, a task and its replicas are processed concurrently by different virtual machines (VMs); the successful completion of any task copy within the deadline triggers a cancellation of all the replicas. More replicas can improve effectiveness of the TRC approach, which however makes the task more vulnerable to co-resident attacks, where a malicious attacker may steal users’ data through co-residing its VMs on the same physical server as users’ VMs. This work solves optimization problems that determine the optimal number of task replicas to minimize the expected user losses, achieving a balance between the task completion probability and the data theft success probability. The solution methodology encompasses a probabilistic model for evaluating the task completion probability by a certain deadline, expected task completion time, and data theft success probability. Examples are presented to demonstrate effects of different parameters (the number of cloud servers, the number of attacker's VMs, and the task deadline) on task performance metrics and optimization solutions.