You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Right now slurm will auto-detect an interface on Linux if there is a single non-loopback inteface. I propose a new heuristic for the case when there are multiple non-loopback interfaces up: slurm should automatically select the interface that has sent/received the most traffic. This information is available via /proc/net/dev, which slurm already has parsing code.
Rationale: many computers actually use a single network interface for all traffic. This isn't necessarily true in HA server environments, but it's probably true for most laptops/desktops and small server instances. The network interface the user cares about the most is probably that network interface. For instance, a common reason that there might be multiple up interfaces is that there are interfaces for VPN connections (e.g. tun0) or interfaces that route to virtual machines on the host (e.g. virbr0). The real physical interface will have at least as much traffic as the VPN interfaces. In most cases the physical interface will also have more data sent than a virtual bridging interface. There are some corner cases where this isn't true, but I think this heuristic will work most of the time.
The text was updated successfully, but these errors were encountered:
Right now slurm will auto-detect an interface on Linux if there is a single non-loopback inteface. I propose a new heuristic for the case when there are multiple non-loopback interfaces up: slurm should automatically select the interface that has sent/received the most traffic. This information is available via
/proc/net/dev
, which slurm already has parsing code.Rationale: many computers actually use a single network interface for all traffic. This isn't necessarily true in HA server environments, but it's probably true for most laptops/desktops and small server instances. The network interface the user cares about the most is probably that network interface. For instance, a common reason that there might be multiple up interfaces is that there are interfaces for VPN connections (e.g.
tun0
) or interfaces that route to virtual machines on the host (e.g.virbr0
). The real physical interface will have at least as much traffic as the VPN interfaces. In most cases the physical interface will also have more data sent than a virtual bridging interface. There are some corner cases where this isn't true, but I think this heuristic will work most of the time.The text was updated successfully, but these errors were encountered: