hachikuji commented on pull request #9401: URL: https://github.com/apache/kafka/pull/9401#issuecomment-729258346
Here are a couple additional test runs. This was on Ubuntu 20 (ami-00831fc7c1e3ddc60). The machine type was m5a.xlarge with 200GB gp2 EBS storage. One instance was running the broker and one instance was running the producer perf test. Commands: ``` bin/kafka-topics.sh --create --topic foo --replication-factor 1 --partitions 10 --bootstrap-server $BROKER bin/kafka-producer-perf-test.sh --topic foo --num-records 250000000 --throughput -1 --record-size 256 --producer-props bootstrap.servers=$BROKER ``` Here are the results: ``` Patch: 250000000 records sent, 826003.925171 records/sec (201.66 MB/sec), 149.39 ms avg latency, 1623.00 ms max latency, 131 ms 50th, 380 ms 95th, 464 ms 99th, 650 ms 99.9th. 250000000 records sent, 825684.740355 records/sec (201.58 MB/sec), 149.22 ms avg latency, 1276.00 ms max latency, 124 ms 50th, 364 ms 95th, 451 ms 99th, 775 ms 99.9th. Trunk: 250000000 records sent, 833144.487250 records/sec (203.40 MB/sec), 148.20 ms avg latency, 1361.00 ms max latency, 111 ms 50th, 437 ms 95th, 551 ms 99th, 807 ms 99.9th. 250000000 records sent, 810927.409022 records/sec (197.98 MB/sec), 152.59 ms avg latency, 1430.00 ms max latency, 127 ms 50th, 382 ms 95th, 467 ms 99th, 809 ms 99.9th. ``` Given variance in these tests, I think we're probably inline with trunk. I looked at the flame graph as well and did not observe any substantial difference in performance. Here are a few interesting highlights from one run. This patch is listed first with trunk second. `Sender.sendProduceRequests`:   `KafkaChannel.write`:   ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
