Modelling And Analysis Of Cloud Platform Based On Erasure Code Storage Mechanism

The technologies of big data and cloud storage have become increasingly mature and been widely applied in various academic and commercial fields recently.Stor-age servers starts to migrate information to cloud.More and more researchers are paying attention to the reliability,security and performance of accessing da ta from cloud platform.The storage utilization and reliability of cloud platform adopted Erasure Code(EC)have been improved significantly comparing to those with tra-ditionally storage mechanism,in which one file are duplicated on multiple server nodes.The performance evaluation model is no longer applicable due to differen-t working strategy.Without a theoretical model the system design,analyze and precise performance evaluation of EC coded cloud platform become hard to accom-plish.An accurate analytical model of cloud platform is vital to system design and performance prediction.Queuing theory model is applied to the server node in EC coded cloud system,the file accessing time is evaluated accurately in such system.In the literature,many researchers adopt M/M/1/K queueing model with single server,finite capacity and Poisson arrival/service processes in which file requests are assumed to arrive to the server node one by one.In the real system,however,file requests may arrive to server node in parallel.The single arrival and service queues M/M/1/K or M/G/1 may not able to describe the real situation precisely.This paper proposes a novel queueing model with batch arrival and service process(MGeo)Through derivation,this paper presents the closed form formulas for the first and second moments of customer waiting time for the MGeo/Meo/1 and MGeo/M/1.By statistics of experimental results,shift exponential distribution(denoted sM)shows advantage than Poisson process when modelling the service process.The z transformation of MGeo/sM/1 queue length distribution and the first and second moments of customer waiting time arc obtained by considering the two Markov chains that are embedded in each batch of customers before and after they arrive.Finally,a strict upper bound of the cloud platform file request delay is proposed,which can be used to predict the upper bound of the file request latency in the EC coded cloud platform.Under different loads of the cloud platform system,the upper bound remains close to the real experimental value.It can be seen that the queuing model and upper bound of file request latency proposed in this paper are of significance in the performance evaluation and prediction of cloud platform system.