Currently we let Linux assign whatever thread to whatever core it wants, but pinning the main thread to a core and then pinning the corresponding worker process to its hyper thread should in theory significantly reduce the latency to communicate with the sandbox. Benchmark the exact difference and, if significant, make it part of the API.
Currently we let Linux assign whatever thread to whatever core it wants, but pinning the main thread to a core and then pinning the corresponding worker process to its hyper thread should in theory significantly reduce the latency to communicate with the sandbox. Benchmark the exact difference and, if significant, make it part of the API.