Quality-Aware Service Selection for Multi-Tenant Service Oriented Systems Based on Combinatorial Auction.
Xuejun Li,Yunxiang Zhong,Qiang He,Feifei Chen,Xuyun Zhang,Wanchun Dou,Yun Yang
DOI: https://doi.org/10.1109/ACCESS.2019.2902131
IF: 3.9
2019-01-01
IEEE Access
Abstract:Multi-tenant service-oriented systems (SOSs) have become a major software engineering paradigm in the cloud environment. Instead of serving a single end-user, a multi-tenant SOS provides multiple tenants with similar and yet customized functionalities and potentially different quality-of-service (QoS) values. Multiple tenants' differentiated multi-dimensional quality constraints for the SOS further complicates the NP-hard problem of quality-aware service selection. Existing quality aware service selection approaches suffer from poor success rates of finding a solution, especially in scenarios where tenants' quality constraints are stringent, due to the lack of systematic consideration of three critical issues: 1) the need to fulfil multiple tenants' differentiated quality constraints; 2) the competition among service providers; and 3) the complementarity between services. This paper proposes a novel approach called combinatorial auction-based service selection for multi-tenant SOSs (CASSMT) to support effective quality-aware service selection for multi-tenant SOSs. CASSMT allows service providers to bid for the services of an SOS expressively. Based on received bids (i.e., QoS offers), CASSMT attempts to find a solution that achieves the system developer's optimization goal while fulfilling all tenants' quality constraints for the SOS. When no solution can be found based on the current bids, service providers can improve their bids to increase their chances of winning, which in the meantime, increases the chances of finding a solution. The experimental results show that CASSMT outperforms representative approaches in the success rate of finding a solution and system optimality. Meanwhile, its efficiency, measured by the number of auction rounds and computation time, is demonstrated to be satisfactory in scenarios on different scales.