Benchmarking

On This Page

This guide walks through a few steps to benchmark Nuclio from scratch.

Setting up a benchmark system

To benchmark Nuclio, you will need three components:

  1. Docker: You’ll use the “local” platform to benchmark so all you’ll need is Docker
  2. wrk: A benchmarking utility
  3. nuctl: All you’ll need is the nuctl CLI. It will pull all the necessary components

Obviously Nuclio will only be as fast as the hardware it runs on. In this case you’ll showcase benchmarks on an AWS c5.9xlarge — a 36 core machine. With Nuclio you leverage parallelism, so adding cores contributes to performance. In these examples you’ll set the # of workers to the # of cores — experiment on your platform to get the most performance.

Benchmark Golang (Go)

Deploy an empty Go function with 36 workers:

nuctl deploy helloworld-go -n nuclio -p https://raw.githubusercontent.com/nuclio/nuclio/development/hack/examples/golang/empty/empty.go --platform local --triggers '{"mh": {"kind": "http", "maxWorkers": 36}}'

Run the benchmark:

wrk -c 36 -t 36 -d 10 http://172.17.0.1:39150
Running 10s test @ http://172.17.0.1:39150
  36 threads and 36 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   105.23us  104.13us  11.52ms   91.01%
    Req/Sec    10.74k   329.50    11.95k    68.97%
  3882355 requests in 10.10s, 336.93MB read
Requests/sec: 384418.84
Transfer/sec:     33.36MB

Benchmark Python 2.7

Deploy an empty Python function with 36 workers:

nuctl deploy helloworld-py -n nuclio -p https://raw.githubusercontent.com/nuclio/nuclio/development/hack/examples/python/empty/empty.py --platform local --triggers '{"mh": {"kind": "http", "maxWorkers": 36}}' --runtime python:2.7 --handler empty:handler

Run the benchmark:

wrk -c 36 -t 36 -d 10 http://172.17.0.1:36724
Running 10s test @ http://172.17.0.1:36724
  36 threads and 36 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     0.93ms    1.04ms  16.76ms   93.78%
    Req/Sec     1.29k   115.26     3.85k    74.71%
  461755 requests in 10.10s, 51.52MB read
Requests/sec:  45718.31
Transfer/sec:      5.10MB

Benchmark PyPy

Deploy an empty Python function with 36 workers:

nuctl deploy helloworld-py -n nuclio -p https://raw.githubusercontent.com/nuclio/nuclio/development/hack/examples/python/empty/empty.py --platform local --triggers '{"mh": {"kind": "http", "maxWorkers": 36}}' --runtime pypy --handler empty:handler

Run the benchmark:

wrk -c 36 -t 36 -d 10 http://172.17.0.1:36724
Running 10s test @ http://172.17.0.1:36724
  36 threads and 36 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.68ms    4.08ms 111.06ms   94.24%
    Req/Sec     1.80k   632.74    12.83k    87.67%
  646945 requests in 10.10s, 56.14MB read
  Socket errors: connect 0, read 2, write 0, timeout 0
Requests/sec:  64056.55
Transfer/sec:      5.56MB

Benchmark .NET Core

Deploy an empty C# function with 36 workers:

nuctl deploy helloworld-dotnetcore -n nuclio -p https://raw.githubusercontent.com/nuclio/nuclio/development/hack/examples/dotnetcore/empty/empty.cs --platform local --triggers '{"mh": {"kind": "http", "maxWorkers": 36}}' --runtime dotnetcore --handler nuclio:empty

Run the benchmark:

wrk -c 36 -t 36 -d 10 http://172.17.0.1:39741
Running 10s test @ http://172.17.0.1:45044
  36 threads and 36 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.22ms    2.78ms  37.28ms   96.15%
    Req/Sec     1.37k   229.02     4.29k    77.13%
  492328 requests in 10.10s, 57.75MB read
Requests/sec:  48746.13
Transfer/sec:      5.72MB

Benchmark Java

Deploy an empty Java function with 36 workers:

nuctl deploy helloworld-java -n nuclio -p https://raw.githubusercontent.com/nuclio/nuclio/development/hack/examples/java/empty/EmptyHandler.java --platform local --triggers '{"mh": {"kind": "http", "maxWorkers": 36}}' --runtime java --handler EmptyHandler

Run the benchmark:

wrk -c 36 -t 36 -d 10 http://172.17.0.1:45906
Running 10s test @ http://172.17.0.1:45906
  36 threads and 36 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.41ms    4.75ms 120.09ms   96.88%
    Req/Sec     1.45k   492.42     2.51k    69.82%
  520925 requests in 10.02s, 58.12MB read
Requests/sec:  51999.28
Transfer/sec:      5.80MB

Benchmark NodeJS

Deploy an empty NodeJS function with 36 workers:

nuctl deploy helloworld-njs -n nuclio -p https://raw.githubusercontent.com/nuclio/nuclio/development/hack/examples/nodejs/empty/empty.js --platform local --triggers '{"mh": {"kind": "http", "maxWorkers": 36}}' --runtime nodejs --handler empty:handler

Run the benchmark:

wrk -c 36 -t 36 -d 10 http://172.17.0.1:39061
Running 10s test @ http://172.17.0.1:39061
  36 threads and 36 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   804.12us    1.08ms  19.78ms   93.50%
    Req/Sec     1.64k   228.41     4.83k    79.54%
  589646 requests in 10.10s, 65.79MB read
Requests/sec:  58384.11
Transfer/sec:      6.51MB