Skip to content

Latest commit

 

History

History
60 lines (47 loc) · 2.09 KB

README.md

File metadata and controls

60 lines (47 loc) · 2.09 KB
Raw

Transport Protocol (Layer 4) Benchmark

This is designed to be a generic network protocol benchmark tailored around specific use cases. It will generate some metrics around the performance of various network protocols under a number of simulated scenarios. As with all benchmarks, take all results with a grain of salt.

Metrics to track:

  • Latency: 1-way, RTT minimum/maximum/mean
  • Throughput: minimum/maximum/mean
  • Jitter: mean
  • Total number of sent/received packets: For reliable protocols, we expect these to be equivalent

Configurable settings:

  • Packet loss - % packets dropped
  • Latency - target latency +- thread sleep resolution
  • Jitter - delay amount to +- to the target latency

Initially, the tests will place some focus on a fast-paced multiplayer game scenario. e.g. FPS games, realtime simulations, etc.

Last thing to note, I'm going to use Rust's default UDP/TCP implementation for those benchmarks.

Protocols to test

Methodology

Each protocol implementation is going to run in a docker container configured using linux's built in Traffic Control (tc) command to simulate unfavorable network conditions. Each test will take place within a single process (multiple threads). The messages themselves will be serialized as JSON to allow for easy inspection. In reality, you would probably want to use something quite a bit more compact.

The follow conditions will be tested using eth0 network interface:

  • 60 hertz send/receive rate

Loss

  • Pristine
  • 10% packet loss
  • 25% packet loss
  • 50% packet loss

Latency

  • 10ms
  • 50ms
  • 200ms

Jitter

  • 0ms
  • 10ms
  • 50ms

Packet Size

  • Less than MTU (~1500 bytes)
  • Greater than MTU