merlimat commented on PR #3383:
URL: https://github.com/apache/bookkeeper/pull/3383#issuecomment-1192777442
@eolivelli @dlg99 @StevenLuMT
We have done a lot of perf tests with this flush-merger enabled in BK and
Pulsar.
1. At low request rates (eg: when there is batching involved), there is
absolutely no visible difference in either the latency or the CPU usage.
2. At high requests rates there is a marked improvement of the latency and
the CPU utilization, due to the reduced number of syscall invocations, and in
the end of bigger TCP frames sent and received.
@StevenLuMT The 1024 number is to increase the chances of bigger
consolidation. It will not affect the latency.
A couple of examples with low-rate scenarios:
Persistent topic:
```
2022-07-04T01:54:25,174+0000 [main] INFO
org.apache.pulsar.testclient.PerformanceProducer -
Throughput produced: 292 msg --- 10.0 msg/s --- 0.1 Mbit/s
--- failure 0.0 msg/s
--- Latency: mean: 2.673 ms - med: 2.652 - 95pct: 3.044 - 99pct:
3.155 -
99.9pct: 3.261 - 99.99pct: 3.261 - Max: 3.261
```
Non-Persistent topic:
```
2022-07-04T01:55:28,361+0000 [main] INFO
org.apache.pulsar.testclient.PerformanceProducer -
Throughput produced: 195 msg --- 10.0 msg/s --- 0.1 Mbit/s
--- failure 0.0 msg/s
--- Latency: mean: 0.703 ms - med: 0.675 - 95pct: 0.978 - 99pct:
1.041 -
99.9pct: 1.241 - 99.99pct: 1.241 - Max: 1.241
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
