Behaviour of redundant scheduler with only one subflow is weird.

[ytxing]

Hi, I am using mptcp v0.95 in my ubuntu 16.04. And I create two lxc container that can communicate sith each other. When I try to download a file from a container, and use redundant scheduler, sudo sysctl net.mptcp.mptcp_scheduler=redundant there is something wrong.

That is, when there is only one available network path (both containers have only one network interface working), the goodput of redundant scheduler is half of the goodput of other schedulers. For example, when I use default scheduler, I got about 600KB/s but with redundant scheduler, it is less than 300KB/s. And ifstat shows that the actual throughput is around 600KB/s so the total bandwidth is the same for all schedulers.

The path manager is fullmesh. Is that normal?

[matttbe]

Hi @ytxing

That's a good question. Maybe this scheduler is doing some restrictive checks that are not needed with only one subflow?

Hopefully @chrpinedo and/or @AlexanderFroemmgen can help there? (I never really looked at this code)

[chrpinedo]

when i worked with this scheduled (years ago), I didn't suffer a behaviour like that.

Te redundant scheduler was designed to replicate the traffic among active paths and load-balancing traffic among backups paths:

• only active paths: redundant traffic through active paths
• only backup paths: load-balanced traffic through backup paths
• mix of active and backup paths: redundant traffic through active paths and backup paths are only used if active paths fail.

So, if you are having the half of the bandwith, could you have 2 active paths through the interface?

This is the only suggestion I can give you. I don't work on this since years.