srl-labs
Prototype separate CPM/Linecard deployment model for VM-based nodes
It is known that distributed nodes (such as most SR OS HW variants, VMx, IOSXRs) are sensitive to the delays between CPM and Linecards. Currently, vrnetlab packs both CPM and LC VMs inside a single container, which makes it impossible to pin CPUs for each of them and thus spread the load more evenly. Eventually, this might cause reboots or datapath issues for such VMs.
The proposal is to prototype a deployment/packaging model where CPM VM is a standalone container, and LC is a standalone container (with respective VMs running inside). At first the networking part might be manually defined in the clab links, but eventually it would make sense to have a node definition that defines a grouping of containers inside of it with networking carried out transparent to the user.
