Extreme HTTP Performance Tuning (talawah.io)

Article summary

The article discusses the process of optimizing HTTP performance on a 4 vCPU AWS EC2 instance, achieving 1.2 million API requests per second. The author outlines the steps taken to optimize the server, including application and framework optimizations, syscall auditing, and interrupt optimizations. The optimizations resulted in a significant increase in performance, from 224,000 requests per second to 1.2 million requests per second. The article also provides a detailed breakdown of the benchmark setup and results.

Main themes

• HTTP performance optimization
• Cloud computing
• Benchmarking
• Latency reduction
• Instance type selection
• NUMA awareness
• Security vs performance tradeoffs
• Cloud infrastructure limits

What commenters say

• Disabling certain CPU power-saving features can improve performance in latency-sensitive applications, but may not be applicable to all instance types.
• Using a load balancer and terminating SSL at the load balancer can improve performance and reduce costs.
• Extreme optimizations may not be directly applicable to real-world applications, but can provide valuable insights into performance bottlenecks.
• The use of specialized instance types, such as Graviton2, can provide significant performance improvements for certain workloads.
• Pinning network workloads to specific NUMA nodes can improve performance, but requires careful consideration of the underlying hardware topology.
• Coordinated omission handling is an important consideration in benchmarking, but may not be relevant in all scenarios.
• Disabling speculative execution mitigations can improve performance, but may also introduce security risks.
• Theoretical maximum request rates for cloud instances are limited by network device limits and quota systems.