Efficient container checkpointing and migration in a decentralized edge orchestration system

With the advent of modern technology, Internet of Things (IoT) has become more commonplace. The widespread usage of IoT-centric technology increases the frequency that new devices will be released into the market, with varying levels of processing power. It is important to ensure lower-end devices, such as edge computing systems, keep up-time parity with higher-end devices and do not fail under high loads. This research aims to address this issue by experimenting with state management via migration and checkpoint creation, reducing stress caused by resource usage on any given system. The presented approach talks about the problems of high resource usage, then elaborates on the migration based solution that it is built off of. This solution, with a decentralized methodology, creates a system where every host can independently manage resources and initiate migrations as needed. The framework of this solution is explained in detail, and how it handles orchestration on edge computing is examined. Then, a checkpoint creation and management system is built on the framework. Ways of stateful checkpoint creation, by way of third party libraries or built-in methods for container providers, are explored. Afterwards, the potential issue of slow transfers for large checkpoints is tackled, by way of creating patches: Lightweight files made by comparing two checkpoints, which can be created non-invasively using system calls provided by the kernel. By using patch files created between consecutive checkpoints during the migration process, an incremental transfer method can be observed. This method is implemented in a decision making process that chooses between patch creation and checkpoint creation. With the data acquired from all the benchmark tests, the research shows the benefits of this new approach to migration.