View in context in the libbitcoin Slack #general mirror.
http client-server benchmark numbers are about to jump.
This was the result of throughput (vs. latency) testing. bs latency is better than bc on all requests except for block (which makes sense, since the whole bc store is just whole blocks on disk, making that faster and everything else slower). But the network architecture of bc limits its ability to deliver those blocks concurrently.
bs doesn't have that limit, but it was hitting a ceiling in throughput at around 16 threads. The stream was the bottleneck, it just wasn't obvious until measuring throughput. But it also creates up to a 54% hit on latency.