Performance evaluation of energy efficient heterogeneous cluster system with PER-CA scheme under docker cloud environment

Abstract

Heterogeneity occurs because server are gradually provisioned and replaced over the typical 15 years lifetime of a data center. Because the power efficiency of each server differs depending on its hardware specification, the scheduler should be aware of this feature to avoid the worst VM instance placement scenarios. To identify feasible cases, we can define objective functions considering these environment but should know the characteristics of all servers, application and even demands to utilize the model. However in reality, it is NP-hard problem because it is very difficult to know all of relations among these arguments, which eventually leads to a lot of overheads. To solve these excessive overheads and find a right server for a new VM, we propose energy efficient heterogeneous system with Power Efficiency Rank based Container Allocation (PER-CA) scheme including (i) Power consumption Per Application (ppA), Resource utilization Per Application (rpA) tables and (ii) the heuristic algorithm for choosing the energy efficient server based on Power Efficient Rank (PER) vectors. The ppA and rpA tables contain the amounts of energy and computing resources per each benchmark application. In addition, all of applications runs in Docker container environment instead of VM instance. The larger number of applications can run on physical machine rather than hypervisor-based virtualization because Docker container supports light-weight virtualization. Our proposed heuristic calculates the similarities among a new demand and data from the ppA and rpA tables and then derives the server lists according to PER in descending order. Referring to the server lists, the scheduler selects the most appropriate physical server in terms of energy saving. The experimental results showed our proposed schemes can deal with a variety of workloads and be more responsive to I/O load pattern than the existing scheme.