Abstract:
Live VM migration has been a powerful tool to
facilitate system maintenance, load balancing, fault
tolerance, and power-saving in data centers. Precopy
technique is the best suited approach for live
migration. Although pre-copy based live migration
provides minimal service downtime, total migration
time is prolonged which affect on the degradation of
VM’s performance. VM needs the improvement in
performance of migration process by reducing the
total migration time. In this paper, working set
prediction using machine learning (WSPML) is
proposed to reduce the total migration time. It uses
the prediction model with historical data during the
live VM migration process. At first, it trains
experimental dataset which includes the performance
parameters collected from various workloads by
machine learning techniques to build the best
prediction model and then predict the working set
which can affect the total migration time. We
evaluated the effectiveness of the working set
prediction algorithm with various workloads with
simulation model and the experimental result shows
that WSPML can more reduce the total migration
time in live VM migration than XEN’s default precopy based live migration.
