Abstract

Recent developments in Cloud Computing have resulted in a new virtualization technology called Docker Containers, which supports services that are placed on machines that share resources of the operating system. Additionally, it is a lightweight virtualization technology that can be deployed on cloud machines. Thousands of Docker containers can be placed to handle millions of user requests. Using Docker architecture, failures are minimized as well. It is very challenging, however, to manage and place various types of containers on cloud machines. Docker Swarm is a solution that supports this challenge. Through it, containers are placed and managed on clusters of machines. Swarm supports a default strategy of container placement called spread strategy, which distributes containers equally across machines in the cluster. Spread strategy does not consider how to best utilize the resources of the docker machine. As a result, load becomes unequal on machines, like some become overloaded and some under loaded. Our main goal is to distribute containers on the docker machine based on how much of its resources are being used. To determine where to place containers on each machine, we consider how much memory is being used. We propose an algorithm for determining which machine has the freest memory and placing a container on that machine. Experimental results are compared between our proposed algorithm and the default spread algorithm. By placing containers on machines in such a way as to prevent overloading or under loading, we have proposed a container placement algorithm. In addition, this strategy minimizes the number of physical machines in a docker cluster