Page History

...

The timeout was set to 200ms. The data size of a request was 100 bytes. For a low timeout case, we chose 75percentile = 60ms and 50percentile = 20. And for a high timeout case, we chose 75percentile = 400ms and 50percentile = 200ms.

The result shows the a dramatic difference in a high enqueue rate area. As the target rate increases, both implementations keep up with the requests initially. However, the old implementation was saturated around 60000 request per second, whereas the proposed implementation didn't show any significant performance degradation in this benchmark.Another thing worth noting is that there was a performance degradation in the current implementation when timeouts happen more often. It may be because the current implementation has to purge the delay queue and watch lists more often. The proposed implementation shows a slight degradation only when the enqueue rate was very high.

...