Performance Analysis of Server Operation with Block Hiring Policy in Cloud Services

In server-hosting service with cloud computing, servers are hired and released depending on the number of jobs in the system. Considering setup cost which is incurred each time new servers are hired, it may save the total cost if one keeps idle servers for shortly-coming future demand instead of returning them once getting idle. In this paper, we consider an operation policy to hire a group of servers instead of hiring one by one. Firstly we present a queuing model to examine the effect of the number of servers handled simultaneously, called block size, on the system performance to consider the optimal server hiring policy. Through numerical experiments, it is found that with the increase in the block size, charging of setup cost gets less frequent as expected but running cost piles up due to more idle servers. It is also observed that mean waiting time and blocking probability get slightly worse with the block size. The performance analysis presented in the paper will be useful to consider the trade-off relationship between performance and cost of cloud services.