Toward integrity assurance of outsourced computing-a game theoretic perspective

摘要

Outsourced computing is gaining popularity in recent years. However, due to the existence of malicious workers in the open outsourced environment, offering high accuracy computing services is critical and challenging. A practical solution for this class of problems is to replicate outsourced tasks and compare the replicated task results, or to verify task results by the outsourcer herself. However, since most outsourced computing services are not free, the portion of tasks to be replicated or verified is restricted by the outsourcer's budget. In this paper, we propose Integrity Assurance Outsourced Computing (IAOC) system, which employs probabilistic task replication, probabilistic task verification and credit management techniques to offer a high accuracy guarantee for the generalized outsourced computing jobs. Based on IAOC system, we perform theoretical analysis and model the behaviors of IAOC system and the attacker as a two-player zero sum game. We propose two algorithms, Interactive Gradient Descent (IGD) algorithm and Tiered Interactive Gradient Descent (TIGD) algorithm that can find the optimal parameter settings under user's accuracy requirement, without or with considering user's budget requirement. We prove that the parameter setting generated by IGD/TIGD algorithm form a Nash Equilibrium, and also suggests an accuracy lower bound. Our experiments show that even in the most severe situation, where the malicious workers dominate the outsourced computing environment, our algorithm is able to find the parameter settings satisfying user's budget and accuracy requirement.