Migrating stateful containers using an autonomous collector
Abstract
With the growth of cloud computing and Infrastructure as a Service, and the need to keep costs low many companies tries to utilize all their available resources as much as possible. Today, there is not currently any solution that allows you to move services based the available resources.
This thesis compares a machine learning approach to create a autonomous collector against using a round robin approach. It creates a solution on how to live migrate containers using these types of selecting what host to visit next, and to select where the containers should be moved to.
Using our solution, we have found a way to visit the servers with the most change on the most, and visit those with the least change less. During our experiments, we have gotten data that supports our algorithm, and solution. In a run of 180 iterations, 71 of the visits where on the host with the most change.