While it is nice that it is faster, ~7 Gb/core-second using a in-process "virtual network", (thus only measuring the protocol implementation itself instead of the rest of the network stack) is not exactly a fast network protocol implementation. That is ~500,000-700,000 full packets per second or ~1.5-2 core-us/packet.
Under those same conditions, you can quite readily do ~100 Gb/core-second (ignoring encryption, encryption will bottleneck you to 30-50 Gb/core-second on modern chips with AES acceleration instructions) in software with feature parity with proper protocol design and implementation.
While it is nice that it is faster, ~7 Gb/core-second using a in-process "virtual network", (thus only measuring the protocol implementation itself instead of the rest of the network stack) is not exactly a fast network protocol implementation. That is ~500,000-700,000 full packets per second or ~1.5-2 core-us/packet.
Under those same conditions, you can quite readily do ~100 Gb/core-second (ignoring encryption, encryption will bottleneck you to 30-50 Gb/core-second on modern chips with AES acceleration instructions) in software with feature parity with proper protocol design and implementation.