Proceedings of ISP RAS


Dynamic Optimization of Workload on Compute Nodes in Private, Public and Hybrid Clouds

A. Chadin (Servionica, Moscow)

Abstract

Cluster's overload is a complex problem and can be solved by using one of stabilization methods: by migrating VMs from one node to another in case of compute nodes. The main goal of this research is to create nova load-balancer service that manages cluster loads and makes decisions of stabilization based on statistics data. This system is designed for automatic workload distribution in a cluster by analyzing workload of compute nodes and migrating VMs from overloaded nodes to underloaded ones. It runs periodically and checks latest statistics to compare with threshold values. If one of the statistics values is greater than the threshold value, the nova load-balancer runs cluster stabilization process to reduce load on cluster by performing live-migration of VMs. Besides workload stabilization, the system also provides an opportunity to reduce power consumption by unloading and suspending underloaded nodes. If statistics values of node are less than underload-threshold values, then all VMs of current node will be migrated to other nodes. When there are no VMs on the node, LB marks this node as suspended and is going to suspended mode (ACPI S3). If nova load-balancer wants to activate suspended nodes to reduce overall cluster load, it uses WOL technology. Workload stabilization in a cluster increases stability and reduces system response time.

Keywords

Balancing; Cluster; Cloud Computing

Edition

Proceedings of the Institute for System Programming, vol. 27, issue 6, 2015, pp. 307-314.

ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).

DOI: 10.15514/ISPRAS-2015-27(6)-19

Full text of the paper in pdf (in Russian) Back to the contents of the volume