bench.md

libp2p-pubsub-benchmark

Benchmarking asyncmachine-go based implementation of libp2p-pubsub from github.com/pancsta/go-libp2p-pubsub.

TestSimpleDiscovery #1

Versions being compared

• origin
• state based (RPC as a channel)

Steps taken in the test

1. Init a pubsub network of {h} hosts
2. Subscribe with all but host0
3. Publish a single message from host0
4. Publish random msgs - {m} per host

Variables

• h:hosts Num. of hosts in the pubsub network
  • 5 to 20
  • step by 5
• m:messages Num. of sent messages (per host)
  • 20 to 100
  • step by 20

Machines

1. PubSub host (20 states)
2. Discovery (10 states)
3. DiscoveryBootstrapFlow (5 states)

Test duration

Created goroutines

Final goroutines

Allocated memory

Memory ceiling

Final memory

GCed memory

Failure rate (states)

Failure rate (origin)

Conclusions

Although the duration and peak memory usage has a reasonable overhead of ~10%, the failure rate needs further looking into, along with final goroutines.

Footer

Version 20240601-v5.0.0 from github.com/pancsta/go-libp2p-pubsub-benchmark