Reliability concept drift online measurement for composite cloud systems

Abstract

Benefiting from the pay-as-you-go business paradigm, the scale of the global cloud computing industry grows rapidly. It has become a possible way to dynamically integrate and cooperate RESTful and SOAP-based component cloud services to develop large-scale complex software systems based on service composition techniques. The constructed composite cloud systems run under dynamic and uncertain operating environments. The issue of how to deal with the concept drift of the reliability data streams to guarantee a stable execution of composite cloud systems has become a grand challenge. To provide some early guidance and act as a trigger mechanism for composite cloud systems reliability adaptation during the system operation and maintenance, we propose a reliability concept drift measurement approach via Sinkhorn distance for composite cloud systems (or RCDMeas) in this paper. This method employs entropy regularization and fixed-point iteration to identify the reliability concept drift between the historical accumulated reliability flow data and the up-to-date reliability flow data. Large-scale data experiments verified the effectiveness of the proposed approach.