Abstract-We model the Virtual Organization (VO) formation problem in grids using concepts from coalitional game theory and design a distributed mechanism for solving it. The proposed distributed mechanism enables the formation of VOs guaranteeing the maximum profit for their participating Grid Service Providers (GSPs). We show that the proposed mechanism produces stable VOs, that is, the GSPs do not have incentives to break away from the current VO and join some other VO. We perform extensive simulation experiments using real workload traces to characterize the properties of the proposed distributed mechanism. The results show that the proposed distributed mechanism not only produces VOs that are stable yielding high revenue for the participating GSPs, but also decides the structure of the VOs in a reasonable amount of time.
Keywords-grid computing; VO formation; coalitional games;I. INTRODUCTION Grid computing systems are composed of heterogeneous resources (CPUs, storage, etc.) owned by autonomous organizations. These systems provide essential resources for conducting cutting-edge science and engineering research. The resource management in such open distributed environments is a very complex problem. One important aspect of resource management in grids is how Grid Service Providers (GSPs) pool their resources together to execute large scale applications. These GSPs collaborate and form Virtual Organizations (VOs) [1].The life-cycle of a VO consists of four phases: (i) identification, in which the possible partners and the VO's objectives are identified; (ii) formation, in which the potential partners (GSPs) negotiate the exact terms, the goal, and the duration of collaboration; (iii) operation, in which the members of the VO collaborate in solving a specific task; and (iv) dissolution, in which the VO is dismantled. This paper focuses on designing mechanisms for the second phase, the formation of VOs. We model the VO formation as a coalitional game where GSPs decide to form VOs in such a way that each GSP maximizes its own profit, the difference between revenue and costs. A GSP will choose to participate in a VO if its profit is not negative. The VOs provide the composite resource needed to execute applications. A VO is traditionally conceived for the sharing of resources, but it can also represent a business model [1]. In this work, a VO is a coalition of GSPs which desire to maximize their individual profits and are largely indifferent about the global welfare. We design a distributed