See relevant envoy-dev thread: https://groups.google.com/forum/#!topic/envoy-dev/33QvlXyBinw

Some potential work here:

• Performance profile different application loads against Envoy running on different numbers of cores.
• Per worker CPU pinning.
• Adding EPOLLEXCLUSIVE and EPOLLROUNDROBIN support to libevent and consume them in Envoy
• Adding per worker stats for connection load, accept rate, etc.
• Investigate Envoy built-in connection rebalance between workers (e.g., allow configurable slew between workers but if some worker gets too far from the highest loaded worker, send a new connection to the other worker).

Note that I've tried variants of ^ over the years and it's extremely difficult to beat the kernel at its own game while generically improving all workloads. With that said, this is a very interesting area to investigate albeit incredibly time consuming.