Samee Ullah Khan

Using game theory for scheduling tasks on multi-core processors for simultaneous optimization of performance and energy

Multi-core processors are beginning to revolutionize the landscape of high-performance computing. In this paper, we address the problem of power-aware scheduling/mapping of tasks onto heterogeneous and homogeneous multi-core processor architectures. The objective of scheduling is to minimize the energy consumption as well as the makespan of computationally intensive problems. The multi- objective optimization problem is not properly handled by conventional approaches that try to maximize a single objective. Our proposed solution is based on game theory. We formulate the problem as a cooperate game. Although we can guarantee the existence of a Bargaining Point in this problem, the classical cooperative game theoretical techniques such as the Nash axiomatic technique cannot be used to identify the Bargaining Point due to low convergence rates and high complexity. Hence, we transform the problem to a max-max-min problem such that it can generate solutions with fast turnaround time.

Comparison and analysis of ten static heuristics-based Internet data replication techniques

This paper compares and analyzes 10 heuristics to solve the fine-grained data replication problem over the Internet. In fine-grained replication, frequently accessed data objects (as opposed to the entire website contents) are replicated onto a set of selected sites so as to minimize the average access time perceived by the end users. The paper presents a unified cost model that captures the minimization of the total object transfer cost in the system, which in turn leads to effective utilization of storage space, replica consistency, fault-tolerance, and load-balancing. The set of heuristics include six A-Star based algorithms, two bin packing algorithms, one greedy and one genetic algorithm. The heuristics are extensively simulated and compared using an experimental test-bed that closely mimics the Internet infrastructure and user access patterns. GT-ITM and Inet topology generators are used to obtain 80 well-defined network topologies based on flat, link distance, power-law and hierarchical transit-stub models. The user access patterns are derived from real access logs collected at the websites of Soccer World Cup 1998 and NASA Kennedy Space Center. The heuristics are evaluated by analyzing the communication cost incurred due to object transfers under the variance of server capacity, object size, read access, write access, number of objects and sites. The main benefit of this study is to facilitate readers with the choice of algorithms that guarantee fast or optimal or both types of solutions. This allows the selection of a particular algorithm to be used in a given scenario.

A cooperative game theoretical replica placement technique

Creating replicas of frequently accessed data objects across a read intensive network can lead to reduced communication cost and end-user response time. On the contrary, data replication in the presence of writes incurs extra cost due to multiple updates. The selection of data objects and servers requires solving a constraint optimization problem, which is NP-complete in general. A majority of the current state-of-the-art replica placement techniques suffer from high computational complexity issues. To circumvent such issues, we propose a cooperative game theoretical technique in which the servers (players) in the system collectively deliberate to converge at a replica schema that is beneficial to the system as a whole. In particular we make use of the Aumann-Shapley mechanism of cooperative game theory to propose an effective replica placement technique that yields good solutions when the system is very heavily loaded. Experimental comparisons are made against: (1) branch and bound, (2) greedy, (3) genetic, (4) Dutch auction, and (5) English auction. As demonstrated by the experimental results, the proposed technique maintains superior solution quality in terms of lower communication cost and reduced execution time.

Passive optical network layout in Manhattan

In this paper, we discuss the passive optical network (PON) deployment on an arbitrary grid. We show that this problem in general is NP-hard. We propose an algorithm, which guarantees a solution of 3-approximation to the optimal deployment, and further argue that this is the best lower bound achievable in our case.

How much more rain

Smart Dust particles, are small smart materials used for generating weather maps. We investigate question of the optimal number of Smart Dust particles necessary for generating precise, computationally feasible and cost effective 3-D weather maps. We also give an optimal matching algorithm for the generalized scenario, when there are N Smart Dust particles and M ground receivers.

Why should we pay more for layout designers

In this paper, we discuss the Passive Optical Network deployment on an arbitrary grid with guaranteed tolerance towards p-1 equipment failure. We show that this problem in general is NP-hard. We propose an algorithm, which guarantees a solution of 4-approximation to the optimal deployment, and further argue that this is the best lower bound achievable in our case. We do comparative studied with randomized layouts, were our proposed algorithm saves 45% - 55% deployment cost (fiber, equipment, etc.) on average.