Evaluation of blind scheduling policies under correlated successive inter-arrival times
MetadataShow full item record
Modeling correlated arrivals of traffic into systems is one of the most challenging tasks. Previous assumptions where arrival of traffic into systems is assumed to follow Poisson distribution have been proved to be inaccurate in the presence of correlated inter-arrival times and therefore the modeling approach becomes unrealistic. In this study, we evaluate blind scheduling policies under correlated successive inter-arrival times of jobs into the system. We model arrival of data traffic into systems using Markov Modulated Poisson Process (MMPP) and simulate using Matlab to observe and analyse system behavior. Performance metric used are mean response time and mean queue length. We used the MMPP/G/1 queue model to evaluate the performance of jobs with correlated and Poisson arrivals under PS, FCFS, and LAS scheduling policies. MMPP captures correlation between inter-arrival times while M/G/1 queue was evaluated for comparison studies. We have investigated the impact of increasing load and job size on the average response time and average queue length. We observed that for the case of PS and LAS scheduling policies, jobs with correlated arrivals perform worse than jobs with Poisson arrivals in terms of both average response time and average queue length. We also observed that under LAS scheduling policy, average response time increases with increase in job size. This trend is also true for average queue length. However, under FCFS, jobs with correlated arrivals perform better than jobs with Poisson arrivals in terms of both average response time and average queue length. We therefore conclude that jobs with correlated arrivals perform better under FCFS policy as compared with PS and LAS policies. Thus, by serving jobs with the right policy in the right order reduces on average response time and average queue length without the need to purchase additional system resources. This model can be adopted by business enterprises and production industries where bursty data traffic can be processed.