Benchmarks

These benchmarks are designed to run off the integration test suite, and provide detailed profiling artifacts generated with pprof.

Prerequisites

To run the benchmarks locally, we take a similar apprach to the integration tests.

Envoy binary envoy available: set ENVOY environment variable to the location of the binary, or use the default value /usr/local/bin/envoy
go-control-plane builds successfully
Local installation of pprof and graphviz for profiler output

To run the benchmark:

make benchmark MODE=0

There are 5 different modes all corresponding to various profiling outputs:

const (
	PPROF_CPU int = iota
	PPROF_HEAP
	PPROF_MUTEX
	PPROF_BLOCK
	PPROF_GOROUTINE
)

To specifiy the mode, use MODE environment variable that corresponds to the output you wish to evaluate.

The output file will be stored at the project root with .pprof extension (e.g. cpu.pprof).

Run pprof tool like so which will open up a shell. From there, you can run command such as web to visualize the result:

go tool pprof cpu.pprof 
(pprof) web

For more information, run help.

To run the benchmarks, we just require the prerequisite of docker and go.

To run the benchmarks:

make docker_benchmarks

This will generate all profile artifacts in the ./benchmarks/reports. Graphical profile anaylsis is located in /.benchmarks/pngs.

For more information on how to interpret these reports/graphs, click here