No abstract

reputation systems are essential to evaluate the trustworthiness of participating peers and to combat the selfish, dishonest, and malicious peer behaviors. The system collects locally-generated peer feedbacks and aggregates them to yield the global reputation scores. Surprisingly, most previous work ignored the distribution of peer feedbacks.We use a trust overlay network (TON) to model the trust relationships among peers. After examining the eBay transaction trace of over 10,000 users, we discovered a power-law distribution in user feedbacks. Our mathematical analysis justifies that power-law distribution is applicable to any dynamically growing P2P systems, either structured or unstructured.We develop a robust and scalable P2P reputation system, PowerTrust, to leverage the power-law feedback characteristics. The PowerTrust system dynamically selects small number of power nodes that are most reputable using a distributed ranking mechanism. By using a lookahead random walk strategy and leveraging the power nodes, the PowerTrust significantly improves in global reputation accuracy and aggregation speed. PowerTrust is adaptable to dynamics in peer joining and leaving and robust to disturbance by malicious peers. Through P2P network simulation experiments, we find significant performance gains in using PowerTrust.This power-law guided reputation system design proves to achieve high query success rate in P2P file-sharing applications. The system also reduces the total job makespan and failure rate in large-scale, parameter-sweeping P2P Grid applications.

No abstract

No abstract

In scheduling a large number of user jobs for parallel execution on an open-resource Grid system, the jobs are subject to system failures or delays caused by infected hardware, software vulnerability, and distrusted security policy. This paper models the risk and insecure conditions in Grid job scheduling. Three risk-resilient strategies, preemptive, replication, and delay-tolerant, are developed to provide security assurance. We propose six risk-resilient scheduling algorithms to assure secure Grid job execution under different risky conditions. We report the simulated Grid performances of these new Grid job scheduling algorithms under the NAS and PSA workloads. The relative performance is measured by the total job makespan, Grid resource utilization, job failure rate, slowdown ratio, replication overhead, etc. In addition to extending from known scheduling heuristics, we developed a new space-time genetic algorithm (STGA) based on faster searching and protected chromosome formation. Our simulation results suggest that, in a wide-area Grid environment, it is more resilient for the global job scheduler to tolerate some job delays instead of resorting to preemption or replication or taking a risk on unreliable resources allocated. We find that delay-tolerant Min-Min and STGA job scheduling have 13-23 percent higher performance than using risky or preemptive or replicated algorithms. The resource overheads for replicated job scheduling are kept at a low 15 percent. The delayed job execution is optimized with a delay factor, which is 20 percent of the total makespan. A Kiviat graph is proposed for demonstrating the quality of Grid computing services. These riskresilient job scheduling schemes can upgrade Grid performance significantly at only a moderate increase in extra resources or scheduling delays in a risky Grid computing environment.

No abstract

Abstract-In this paper, we present generic cloud performance models for evaluating Iaas, PaaS, SaaS, and mashup or hybrid clouds. We test clouds with real-life benchmark programs and propose some new performance metrics. Our benchmark experiments are conducted mainly on IaaS cloud platforms over scaleout and scale-up workloads. Cloud benchmarking results are analyzed with the efficiency, elasticity, QoS, productivity, and scalability of cloud performance. Five cloud benchmarks were tested on Amazon IaaS EC2 cloud: namely YCSB, CloudSuite, HiBench, BenchClouds, and TPC-W. To satisfy production services, the choice of scale-up or scale-out solutions should be made primarily by the workload patterns and resources utilization rates required. Scaling-out machine instances have much lower overhead than those experienced in scale-up experiments. However, scaling up is found more cost-effective in sustaining heavier workload. The cloud productivity is greatly attributed to system elasticity, efficiency, QoS and scalability. We find that autoscaling is easy to implement but tends to over provision the resources. Lower resource utilization rate may result from auto-scaling, compared with using scale-out or scale-up strategies. We also demonstrate that the proposed cloud performance models are applicable to evaluate PaaS, SaaS and hybrid clouds as well.

Abstract-As cloud computing becomes more and more popular, understanding the economics of cloud computing becomes critically important. To maximize the profit, a service provider should understand both service charges and business costs, and how they are determined by the characteristics of the applications and the configuration of a multiserver system. The problem of optimal multiserver configuration for profit maximization in a cloud computing environment is studied. Our pricing model takes such factors into considerations as the amount of a service, the workload of an application environment, the configuration of a multiserver system, the service level agreement, the satisfaction of a consumer, the quality of a service, the penalty of a low quality service, the cost of renting, the cost of energy consumption, and a service provider's margin and profit. Our approach is to treat a multiserver system as an M/M/m queueing model, such that our optimization problem can be formulated and solved analytically. Two server speed and power consumption models are considered, namely, the idle-speed model and the constantspeed model. The probability density function of the waiting time of a newly arrived service request is derived. The expected service charge to a service request is calculated. The expected net business gain in one unit of time is obtained. Numerical calculations of the optimal server size and the optimal server speed are demonstrated.