Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Following is throughput difference among 1, 5, 10 max.in.flight.requests.per.connection. The first table uses a server is in AWS us-east-1 and the a producer is in AWS ap-northeast-1.

max.in.flight.requests.per.connectionthroughputavg latencymax latency50th latency95th latency99th latency99.9th latency
119.623234 records/sec25093.64 ms49788.00 ms25193 ms47286 ms49341 ms49788 ms
595.129376 records/sec5111.89 ms10096.00 ms5135 ms9644 ms10095 ms10096 ms
10 (temporary change hard code value in producer and broker)183.083120 records/sec2558.52 ms5036.00 ms2528 ms4749 ms5034 ms5036 ms

The second table uses server and producer via localhost. As the result, more in-flight requests still improve throughput in low latency environment.

./bin/kafka-producer-perf-test.sh --topic quickstart-events --num-records 10000000 --throughput 1000000000 --record-size 1000 --transactional-id thisistest --bootstrap-server 127.0.0.1:9092 --command-property max.in.flight.requests.per.connection=x

max.in.flight.requests.per.connectionthroughputavg latencymax latency50th latency95th latency99th latency99.9th latency
1198562.408165 records/sec (189.36 MB/sec)126.01 ms464.00 ms147 ms171 ms260 ms430 ms
5285087.094107 records/sec (271.88 MB/sec)2.81 ms127.00 ms0 ms14 ms68 ms108 ms
10 (temporary change hard code value in producer and broker)307238.540002 records/sec (293.01 MB/sec)2.14 ms76.00 ms0 ms8 ms45 ms73 ms

Public Interfaces

Broker Configuration

...