Some practical aspects in modeling server clusters with blocks of reserve servers

Abstract

Energy consumption accounts for a significant portion of a server cluster's budget. This has resulted in much interest being generated over the years in how to most effectively reduce these energy consumption costs. Unfortunately, many researchers have omitted some practical aspects with regard to the operation of servers. Thus, they may not have adequate facts for calculating the amount of electrical energy consumed by servers. This paper has investigated the impact of some practical aspects in the operation of servers on calculating the time that the servers consume energy. The considered aspects are S1) the avoidance of turning off the server during the setup process, S2) the separate switching times required for each server, and S3) the necessity of allowing for the shutdown time of each server. The results show that, the assumption for turning off the server during its setup process should be carefully applied in modeling the server cluster. This aspect shows the considerable impact of energy consumption on the operative servers and affects the length of time that energy is consumed in the case of the mean service time is small enough compared to the mean switching times. Additionally, we might omit the shutdown time while still keeping the accuracy of models.