mmmries

defmodule Benchmarks do
  def init do
    :ok = :lbm_kv.create(Web.Job)
  end

  def count_entries do
    IO.puts "Web.Job => #{:lbm_kv.match_key(Web.Job, :_) |> elem(1) |> Enum.count}"
  end

  def measure_throughput(fun, num_items) do

mmmries

Keybase proof

I hereby claim:

• I am mmmries on github.
• I am mmmries (https://keybase.io/mmmries) on keybase.
• I have a public key whose fingerprint is 99B1 CC8A FE86 779F 3F7C 3D80 DA6C FBE7 2629 D9CF

To claim this, I am signing this object:

mmmries

quick_benchmark = fn(tcfn, tcn) ->
  tc_l = :lists.seq(1,tcn) |> Enum.map(fn(_)-> tcfn |> :timer.tc |> elem(0) end)
  tc_min = :lists.min(tc_l)
  tc_max = :lists.max(tc_l)
  tc_med = :lists.nth(round((tcn - 1) / 2), :lists.sort(tc_l))
  tc_avg = round(Enum.sum(tc_l) / tcn)
  %{min: tc_min, max: tc_max, median: tc_med, average: tc_avg}
end

# quick_benchmark.(fn() -> do_some_work() end, 1000) will do 1000 iterations of work and report median, average, min and max back to you in microseconds

mmmries

Protobuf Benchmarking

A really basic protobuf benchmark comparing the protobuf encoding/decoding performance. For details please see the benchmark.rb file in this gist and check the gpb benchmark files.

Latest Results

• protobuf-3.6.12
• gpb-3.26.6

mmmries

Phoenix Nodes

First I created 3 droplets on digital ocean with 4-cores and 8GB of RAM. Login as root to each and run:

sysctl -w fs.file-max=12000500
sysctl -w fs.nr_open=20000500
ulimit -n 4000000
sysctl -w net.ipv4.tcp_mem='10000000 10000000 10000000'

mmmries

defmodule LatencyBenchmark do
  @default_settings %{
    num_actors: 1,
    actions_per_actor: 1,
  }

  def benchmark(action_fn, settings) do # = num_actors, actions_per_actor, setup_fn, action_fn) do
    settings = Map.merge(@default_settings, %{setup_fn: fn -> %{} end}) |> Map.merge(settings)
    settings = Map.put(settings, :action_fn, action_fn)
    {:ok, collector_pid} = Agent.start_link(fn -> [] end)

mmmries

mmmries

The main idea here is compose shared functionality into a pipeline of functions that all implement some shared behaviour.

defmodule Notification.Event do
  # The event (probably a bad name) is where you would put the structified JSON event you got from RabbitMQ
  defstruct [:sent_at, :user, :event]
end

defmodule Notification.Plug do
  @callback init(opts) :: opts

mmmries

• Testing
  • mocking hardware
  • what is worth testing?
• Dev environment vs firmware
• Sensors:
  • Moisture Sensors
  • Temperator Sensors
  • Magnetic Sensors
• Touch Screen
• Carry the touch screen from device-to-device

mmmries

I wanted to run another round of performance benchmarks for gnat to see how it's request throughput has changed with the introduction of the ConsumerSupervisor which handles things like processing each request in its own supervised process.

I used a CPU-optimized digital ocean droplet with 16 cores, gnatsd 1.3.0, erlang 21.2.2 and elixir 1.8.0.rc0 You can read the setup instructions below for more details and the results_by_concurrency.md contains details about a lot of different runs.

I'm trying to measure the overhead in the system, so the requests are random byte strings that just get echoed back with processing. The measurements use byte strings of 4 bytes up to 1024 bytes.

TL/DR; You can do 170k+ synchronous requests/sec of small messages or 192MB+/sec of 1kb messages