nice work! we have handrolled something similar at work.

do you have any perf metrics? throughput, end-to-end latency, etc?