Swarm2k and Swarm3k were collaborative experiments. We raised funds in terms of Docker Hosts, instead of money, with a call to participate. The result was astonishing-Swarm2k and Swarm3k were joined by dozens of individuals and corporate geographically distributed contributors. In total, for Swarm2k, we collected around 2,300 nodes, while for Swarm3k, around 4,700.
Let's discuss the architecture of Swarm2k. In the preceding figure, there are three managers, denoted as mg0, mg1, and mg2. We will use the three managers because it is the optimum number of managers, suggested by the Docker core team. Managers formed a quorum on a high-speed network link and raft nodes employ employed a significant amount of resources to synchronize their activities. So, we decided to deploy our managers on a 40GB Ethernet link into the same data center.
At the beginning of the experiment, we had the following configuration:
mg0 was the cluster's manager leader
mg1 hosted the stat collector...