Attention is currently required from: jolly, pespin.
1 comment:
Patchset:
Several comments: […]
I think for now it's good to have it optional so we can compare, indeed. The delayed submit will likely increase throughput - but at the cost of some added latency. I think it's worth it, but before we've done extensive benchmarks (and asked users to try it in loaded production setups) it might be good to have it disabled by default. As we don't have VTY in all libosmocore applications, and IO_URING is enabled via environment, adding another env variable is fine with me.
But I agree with @pespin regarding the global variable. Note that it will likely have to be __thread / thread-local-storage like all of our I/O. It should be very chaep to set and check compared to unneeded io_uring_submit() all the time without knowig if we actually did submit any new sqe during that given select loop iteration.
To view, visit change 40489. To unsubscribe, or for help writing mail filters, visit settings.