IJCA, Vol. 20, No. 1, March 2013 ISCA Copyright© 2013 1 An Economic Model for Resource Adaptation in 2D Mesh Multicomputer Networks Ayoub Alsarhan * , Emad E. Abdallah * Ibrahim Al-Oqily † , and Alaa Eddien Abdallah † Hashemite University, Zarqa, JORDAN Abstract In the current partitionable Mesh-connected processors parallel machine, the system administrator (SA) can provide a means for offering the required quality of service (QoS) for a multiple classes of users. We consider the approach where the SA rents free processors in a 2D mesh-connected multicomputer with QoS guarantees for the users. Free processors are managed by the SA resource management policy. SA policy should provide QoS for the users. Clearly, maximizing profit is the key objective for the SA. In this work, we propose a novel resource adaptation approach which uses an economic model to derive processors renting policy that can adapt to changing system load and users demand for services. Our scheme uses an economic model for trading computing resources. The economic model includes costs and revenues of serving users’ requests. The main concern of the derived resource management scheme is to allow continuous optimizing of the SA profit, while keeping acceptable Grade of Service. The approach integrates computing resource adaptation with service admission control based on Markov Decision Process theory. Numerical analysis stresses the ability of our approach to maximize SA profit under varying system conditions. Key Words: Resource management, processors allocation, resource management, 2D mesh connected multicomputer network, partitionable parallel machine. 1 Introduction Mesh-connected processors are distributed computing infrastructure which consist of many independent processors connected through a high-speed network [8-9, 13-14]. It is used to solve large-scale computation problems. The machine can be divided into a set of clusters where the processors in a cluster are used to execute parallel jobs [13-14]. The computational resources in a parallel machine are managed by SA. SA tries to execute users’ jobs by allocating free processors to the users’ requests. SA should be able to assign the jobs efficiently from various users to the free processors in * Department of Computer Information System. Email: AyoubM@hu.edu.jo and Emad@hu.edu.jo. † Department of Computer Science and Application. Email: izaloqily@hu.edu.jo and aabdallah@hu.edu.jo. a 2D Mesh machine which is commonly known as processors allocation problem [8-9]. The purpose of processors allocation in our model is to maximize SA profit. SA charges users in order to execute their parallel tasks. The following are key objectives for the SA: • Providing enough QoS for users. This objective can be achieved by offering an adequate number of processors for serving users requests. For this purpose, processors with guaranteed QoS are leased by the users from the SA using Service Level Agreements (SLA). SA serves new requests on the basis of a Request Admission Control (RAC) policy that ensures request effective processors availability. • System Grade of Service (GoS). The RAC policy will block the number of requests to ensure good GoS to users. Requests blocking probabilities must be constrained to acceptable values. • SA profit which is basically defined as reward minus costs. This objective aims to optimize SA profit, taking into account the cost of renting processors. For the SA, the rewards are the real revenue from serving users requests. In our proposed approach, users pay for individual requests. Hence, SA’s rewards are computed using the amount of admitted requests. It is widely known that system load fluctuates and changes over time and thus optimal static solution is inadequate to solve this problem especially in cases of system overload, and inefficient network resources use in case of network underload. In this paper, we propose a new approach based on an economic model. The model realizes the continuous profit optimization for the SA. It also integrates the adaptation of the offered number of processors (to traffic load and renting cost variations) with the RAC. To satisfy GoS, the approach adapts the number of allocated processors for each class of users to continuously meet request blocking constraints. Our trading approach can be applied in a style of computing where users can access information services. SAs trade their services on cloud resources for money. SA may trade any resource from infrastructure such as processors, bandwidth, network, platforms and applications. The major contributions of this paper are summarized as follows: (1) A model for renting processors in the 2D mesh