A machine learning-based optimization approach for pre-copy live virtual machine migration
Raseena M. Haris,Khaled M. Khan,Armstrong Nhlabatsi,Mahmoud Barhamgi
DOI: https://doi.org/10.1007/s10586-023-04001-1
2023-05-11
Cluster Computing
Abstract:Organizations widely use cloud computing to outsource their computing needs. One crucial issue of cloud computing is that services must be available to clients at all times. However, the cloud services may be temporarily unavailable due to maintenance of the cloud infrastructure, load balancing of services, defense against cyber attacks, power management, proactive fault tolerance, or resource usage. The unavailability of cloud services impacts negatively on the business model of cloud providers. One solution to tackle the service unavailability is Live Virtual Machine Migration (LVM), that is, moving virtual machines (VMs) from the source host machine to the destination host without disrupting the running application. Pre-copy memory migration is a common LVM approach used in most networked systems such as the cloud. The main difficulty with this approach is the high rate of frequently updating memory pages, referred to as "dirty pages. Transferring these updated or dirty pages during the pre-copy migration approach prolongs the total migration time. After a predefined iteration, the pre-copy approach enters the stop-and-copy phase and transfers the remaining memory pages. If the remaining pages are huge, the downtime or service unavailability will be very high -resulting in a negative impact on the availability of the running services. To minimize such service downtime, it is critical to find an optimal time to migrate a virtual machine in the pre-copy approach. To address the issue, this paper proposes a machine learning-based method to optimize pre-copy migration. It has mainly three stages (i) Feature selection (ii) Model generation and (iii) Application of the proposed model in pre-copy migration. The experiment results show that our proposed model outperforms other machine learning models in terms of prediction accuracy and it significantly reduces downtime or service unavailability during the migration process.
computer science, information systems, theory & methods