|
|
|
|
|
|
by throwaw12
332 days ago
|
|
|
> found that it was mostly measuring client implementation latencies and other internal inefficiencies... As you said benchmarking is hard, but isn't the end to end latency customers will see in their workloads is usually including the client library overheads? IMO, benchmarks should closely resemble the real world scenarios (excluding variables, e.g. network latency of different cloud providers) |
|
|