Currently, ECCL lacks a mechanism for local inter-process GPU communication over Unix domain sockets. This…
Currently, ECCL lacks a mechanism for local inter-process GPU communication over Unix domain sockets. This…: a task in LegoFlow-SWE (Harbor dataset). The project needs a UdsExchanger that lets a rank wait for a peer with accept from(peer rank, timeout ms) and another rank connect with connect…
The task
The project needs a `UdsExchanger` that lets a rank wait for a peer with `accept_from(peer_rank, timeout_ms)` and another rank connect with `connect_to(peer_rank, timeout_ms)`. Both methods should return `true` on success and `false` if the operation cannot finish before the given timeout. The exchanger must tolerate…
Part of Lego-X/LegoFlow-SWE.