Or p98 time for requests. Throughput and latency are usually at odds with each other.