Research On Elastic Scheduling In Cloud Data Center Network

As a new computing service model that supports commercial computer clusters to process large volumetric data,cloud computing refers to both applications provided through the Internet as services and hardware and system software in data centers that provide these services.With the exponential growth of Internet services,the number and size of cloud services have increased dramatically in many application areas.The primary goal of the cloud service is to rapidly provision and publish configurable computing resources(eg.networks,servers,storage,applications,and services with minimal administrative effort or service provider interaction)as a shared pool.As an irreplaceable key infrastructure driving this growing trend,data center networks need to be able to respond quickly to users' business requests and achieve efficient network resource management.This paper studies the problem of elastic scheduling of cloud data centers in the process of resource allocation.The specific research work mainly focuses on maximizing the elasticity of the cloud data centers under different constraints.In the research process of various aspects,new optimization models are constructed according to different constraints,and corresponding solutions are proposed.Then,we compare the solutions based on the simulations and real data center platform with the existing methods to verify the feasibility of the theoretical model and the correctness of the solution.The specific research work includes the following three aspects:(1)Firstly,we study the elastic resource scheduling problem of cloud data center under the hose model.It means that given the cloud data center topology and virtual cluster requests,we try to find an elastic allocation in the cloud data center with hose model by adopting a reasonable resource scheduling scheme.The objective of our study is to find a configuration that will support future growth maximization without load redistribution during the runtime.We first propose a distributed linear optimal solution based on message passing under the tree-structured data center network,and we discuss several properties and extensions of the model.Based on the assumptions and conclusions,we extend it to the multiple paths case with a Fat-tree data center,and we discuss the optimal solution for computing the maximum admissible load with both computation and communication constraints.After that,we present the provisioning scheme with the maximum elasticity for the virtual machines,which comes with provable optimality guarantee for a fixed flow scheduling strategy in a Fat-tree data center network.We conduct the evaluations on our test-bed and present various simulation results by comparing the proposed maximum elastic scheduling schemes with other methods.IV Extensive simulations validate the effectiveness of the proposed policies,and the results are shown from different perspectives to provide solutions based on our research.(2)Secondly,we consider elastic scaling for existing virtual clusters to maximize elasticity with the constraint of communication cost in the data center network.It means that given the distribution of users in a cloud data center and the status of resource scaling requests,we try to realize the cloud data center by adopting a reasonable resource scheduling scheme according to the maximum growth limit of the communication cost.Our objective is also to maximize the elasticity during the resource scheduling process.According to the different characteristics of the users' requests,we divide the problem into the following three cases and give corresponding resource allocation strategies with the goal of maximizing the elasticity.a)Single virtual cluster scaling.Firstly,we consider the single virtual cluster scaling problem under the communication cost,and we propose a provable optimal elastic resource scheduling method based on virtual cluster location awareness.b)Multiple virtual clusters scaling.After that,we extend our scheme for the scaling of offline multiple virtual clusters under the communication cost.We consider the elastic resource scheduling for a group of users in a period of time.We prove that multiple virtual clusters scaling for the over-time elasticity maximization problem is NP-hard,and we propose a heuristic method for the virtual clusters.c)Online multiple virtual clusters scaling.Finally,we consider the elastic resource scheduling problem under the online case.Two heuristic algorithms are proposed by using Bayesian parameter estimation to solve an online scaling with both synchronous and asynchronous incoming rates.Extensive simulations demonstrate that our elastic virtual cluster scaling placement schemes outperform existing state-of-the-art methods in terms of flexibility in the data center network.(3)Finally,we study the multiple virtual cluster placement problem with the hose model under the computation and communication constraints.Based on the research work above,we increase the number of users that request at the same time and remove the constraints on the communication cost.We use the notion of elasticity to measure the potential growth of multiple users in terms of both computation and communication resources and our objective is to maximize the elasticity for data center networks.We first formulate this problem as an Integer Linear Programming problem.However,the formulated Integer Linear Programming problem cannot be solved by the simplex or eclipse methods because of a large number of variables and constraints.Therefore,we propose an efficient scheme based on Dynamic Programming and analyze its optimality and complexity.Furthermore,we propose a heuristic algorithm for placement that improves the elasticity and guarantees the bandwidth demand as well as lower complexity.Extensive evaluations demonstrate that our schemes outperform existing state-of-the-art methods in terms of both elasticity and efficiency.