gcool-info / lightning-benchmark

Node benchmark

This repository contains a load test for lightning node software.

The test creates a setup with a bitcoind regtest node and two connected lightning nodes instances.

A test application spins up 100 workers that continously request an invoice from one instance and pay it from the other one.

Output is the number of transaction that are settled per second (TPS) and the average payment latency.

How to run

./run.sh <configuration>

The following configurations are available:

Configuration	Implementation	Backend	Channels	Workers	Options
lnd-bbolt	lnd 0.12.1	bbolt	10	100
lnd-bbolt-keysend	lnd 0.12.1	bbolt	10	100	keysend
lnd-etcd	lnd 0.12.1	single etcd instance	10	100
lnd-etcd-cluster	lnd 0.12.1	three-instance etcd cluster	10	100
clightning	c-lightning 0.9.3	sqlite	1 [1]	100
eclair	eclair 0.6.1	sqlite	10	100
eclair-postgres	eclair 0.6.1	postgres	10	100

¹ Multiple channels are not supported in c-lightning

Results

Below are the test results after 10,000 payments on the following machine:

• Google Cloud n2d-standard-8 instance (8 vCPUs, 32 GB memory)
• 100 GB zonal pd-ssd with ext4 filesystem
• Ubuntu 20.04 LTS

Configuration	Transactions / sec	Avg latency (sec)
eclair	89	1.1
eclair-postgres	46	2.1
clightning	61	1.6
lnd-bbolt-keysend	35	2.8
lnd-bbolt	33	3.0
lnd-etcd	4	29.2
lnd-etcd-cluster	4	31.8

Profiling

For lnd nodes, a cpu profile can be extracted for further analysis. The sender node profiler is reachable through port 5000 on the host. The receiver node profiler is available at port 5001.

To display the profile in a browser, run:

go tool pprof -http 0.0.0.0:7777 http://localhost:5000/debug/pprof/profile

Configuration

The loadtest container reads test parameters from the file loadtest.yml the following parameters are available:

• paymentAmountMsat: the test amount that is paid
• processes: the number of parallel processes
• channels: the number of channels between the two test nodes
• channelCapacitySat: capacity of the channel(s)