# Optimizing Kafka Throughput at Human Security | Adrian Woodhead | Voices of the Kafka Community

[Voices of the Kafka Community](https://www.conduktor.io/voices-of-the-kafka-community)

Optimizing Kafka for Throughput

Guest
Adrian Woodhead
Distinguished Engineer, Human Security

Host
[Ron Kapoor](https://www.conduktor.io/authors/ron-kapoor)
Technical Product Marketing, Conduktor

Adrian Woodhead is a Distinguished Engineer at Human Security, where bot-detection signal data from across the web flows through Kafka to decouple their front-end and back-end systems. He wrote up the throughput experiment in this conversation on the [Human Security engineering blog](https://www.humansecurity.com/tech-engineering-blog/optimizing-kafka-for-throughput/).

In this conversation

→How Human Security uses Kafka to decouple bot-detection signal data
→Building a load-test harness to emulate production traffic
→Why gzip pinned CPU through thread contention under load
→How switching to the less efficient snappy raised throughput
→The benchmarking discipline that made the result trustworthy

"Snappy is less efficient than gzip, but it uses the CPU less. And we got much higher throughput."
Adrian Woodhead, Distinguished Engineer at Human Security

[← More Voices of the Kafka Community](https://www.conduktor.io/voices-of-the-kafka-community)