Clustering

Since version 2.4.0, Cisco Modeling Labs has offered a clustering feature.

When you deploy a CML cluster, the running simulation for a lab is no longer restricted to the resources of a single computer (the all-in-one controller). It can instead use the resources from multiple servers.

Note

Clustering is only available in the non-personal CML product offerings, such as CML-Enterprise and CML-Education.

Clustering Terminology

The documentation uses the following terms when referring to a CML cluster:

node / VM

a virtual machine, which is part of a lab topology

fabric

the software component that connects VM NICs with each other, also provides external connectivity

controller

the CML instance that hosts the CML controller services, the CML UI and the CML APIs

compute or compute member

a member of a cluster where CML runs the VMs for running nodes

cluster

a set of CML servers with one controller and one or more compute members

You can configure the CML controller in a cluster so that it runs VMs just like the compute members, or you can use a dedicated controller that does not run any node VMs. We generally recommend disallowing VMs from running on the controller, if possible.

Clustering Overview

When planning your CML deployment, you have two ways to scale your CML server.

  1. Vertical scaling - increase CML’s capacity by provisioning a larger CML server or adding more and faster CPU cores, RAM, and disk storage to a single server.

  2. Horizontal scaling - increase CML’s capacity by adding additional servers to a CML cluster.

CML clustering adds the ability to connect one or more computes to a CML controller, forming a CML cluster. Within such a cluster, users can start large topologies, which would not fit into a single all-in-one CML server, and the CML controller will dynamically assign nodes to the compute members of the cluster. The cluster controller can place VMs for nodes from a single topology on different computes. During the initial setup of a CML instance, you must choose whether it will be a cluster controller, a cluster compute, or a standalone (non-cluster) instance.

Clustering provides the following high-level features:

  • Dynamic registration of compute members to the cluster controller

  • Sharing of reference platform images from the cluster controller to each compute member

  • Automatic distribution of large labs across available compute members

  • In-place upgrade from a standalone CML server to a CML cluster where the controller is also permitted to run node VMs.

Clustering constraints

Before deploying a CML cluster, you should be aware of the current constraints. In this release, CML clustering has the following constraints:

  • Clustering is only available on non-personal CML products. Personal products, i.e., CML-Personal and CML-Personal Plus, cannot run a CML cluster.

  • No influence on placement. Nodes are placed in no particular order on computes. CML starts the node VM on an arbitrary compute that can run it.

  • No migration of nodes. Once a node has been allocated to a specific compute, it will stay on that compute every time you start the lab unless you wipe the node (or lab) first. The next time you start a node after wiping it, it will be assigned to a compute again. It may be allocated to the same compute as before.

  • No re-balancing. For example, if all running nodes were allocated to one compute, and other computes of the cluster are idle, there is no automatic re-balancing of the nodes to the available computes. As above, stopping and wiping the lab will cause the nodes to be reallocated the next time you start the lab, and that reallocation may rebalance the nodes across the compute members.

  • No external connectivity via computes. All external traffic goes through the controller.

  • Software upgrade for a cluster. To upgrade a cluster, you must manually trigger the upgrade of the controller and all of the computes.

  • A CML cluster may be licensed for up to 300 additional nodes. That is, you may add an 300 expansion node licenses on top of the CML base license.

  • Cluster size is currently limited to 8 computes.