Control plane and data plane

PDFRSS

AWS separates most services into the concepts of control plane and data plane. These terms come from the world of networking, specifically routers. The router's data plane, which is its main function, is moving packets around based on rules. But the routing policies have to be created and distributed from somewhere, and that's where the control plane comes in.

Control planes for your cell-based architecture provide the administrative APIs used to provision, move, migrate, update, remove, deploy, and monitor cells, among others. The data plane is what provides the primary function of the service together with cell router.

To understand the relationship between the control plane and data plane in a cell-based architecture, imagine that you have five cells and the number of users starts growing. Your control plane is responsible for provisioning a new cell and letting the router to know where traffic needs to be sent to. After that, both the router and the cell will be just performing the work they're supposed to (data plane).

Another important aspect to consider here is the static stability as recommended in the Well-Architected Framework, REL11-BP04 Rely on the data plane and not the control plane during recovery. In a statically stable design, the overall system keeps working even when a dependency becomes impaired. For the cell-based context, it would be the data plane to continue operating even if the control plane is down, or even if some availability zone is also down.

Control planes are statistically more likely to fail than data planes. Although the data plane typically depends on data that arrives from the control plane, the data plane maintains its existing state and continues working even in the face of control plane impairment. Data plane access to resources, once provisioned, has no dependency on the control plane, and therefore is not affected by any control plane impairment. In other words, even if the ability to create, modify, or delete resources is impaired, existing resources remain available. You can implement different patterns to be statically stable against different types of dependency failures depending on your cell design strategy.

In addition, we can say that control planes are designed to fail rather than corrupt or provide incorrect information (CP in the CAP theorem), while data planes generally prefer AP in the CAP theorem (Data planes try their best to remain available, even if they depend on stale information for decisions.) An example would be Using a compute layer like Amazon EC2, Amazon ECS or Amazon EKS and Amazon S3 for cell mapping as cell router described later in this paper. Routes to cells are loaded into memory from an S3 bucket. Even if the control plane, Amazon S3 or a zone is unavailable, the router is still able to direct traffic to the cells.