SaRa: A Stochastic Model to Estimate Reliability of Edge Resources in Volunteer Cloud

Abstract

With the increasing popularity and the need for low-cost green computing systems, new paradigms and models such as fog, edge, and volunteer cloud computing (e.g. cuCloud) have recently emerged. cuCloud, one of the appealing volunteer cloud computing system, share the same philosophy as desktop grid, which runs on underutilized and or spare resources of personal computers (i.e. volunteer hosts) owned by individuals and organizations. On one side of the spectrum, underlying cuCloud infrastructure comprises varying levels of availability, volatility, and trust, allowing volunteers to randomly join and leave the model, which makes the resource management and scheduling of tasks a challenging process. On the other side, it is even more challenging and critical to guarantee the Quality of Service (QoS) for applications deployed in the cuCloud model, which requires the tracking and monitoring the reliability and trust of highly distributed volunteer resources. The majority of the available reputation models consider only the ratio of successfully completed tasks to total tasks requested in the determination of reliability decisions of the volunteer nodes, which, in turn, make the reliability model coarse-grained. These models lack of fine-grained parameters such as task-level behaviors (e.g. success or fail) and task characteristics (e.g. priority of a task). To address these challenges, we propose SaRa, a probabilistic system to estimate the reliability of untrusted edge resources in volunteer cloud. Our validation results showed that SaRa's reputation model obtained better reliability estimation than existing methods.