Skip to content

Latest commit

 

History

History
70 lines (57 loc) · 2.12 KB

recap.md

File metadata and controls

70 lines (57 loc) · 2.12 KB
Raw

MQTT-benchmarking

Goal

Analyze the behaviour and the performances of four MQTT brokers among the main ones, plus MQTT-ST (mosquitto bridging).

Behaviour

  • How the cluster is created (number of messages, traffic in byte)
  • How/if connections are forwarded
  • How/if subscriptions are forwarded
  • How messages are forwarded
    • flooding to everyone
    • chain
    • tree

Questions about behaviour

EMQX

  • connect/disconnect: forward connection to everyone (no multiple clients with the same ID)
  • subscripton/unsubscription: forward subscription to everyone
  • publish: send only to the intereset ones (check!!)

RABBITMQ

VERNEMQ

HIVEMQ

If emqx and the other brokers store informations about subscriptions and connection, I'd expect an increase of RAM with the increase of clients and number of topics.

If they forward publications only to the interested ones, I'd expected a low traffic inside the broker. Otherwise -forward to everyone-, an high traffic.

Check subscription forward with multi-level topics

Message forwarding: follow a chain/tree or broadcast to everyone

Performance

  • Network

    • end-to-end delay: time from pubblication to subscription
    • out throughput: how many messages can send a broker
    • in throughput (emitted from the client)
    • total bytes in the cluster (tcpdump during every simulation)

  • Hardware

    • CPU
    • RAM

Experiments

  1. Create the cluster: done
  2. Create N clients: 1 docker with N clients or a docker for each client
  3. Connect N clients:
    • 100% locality: pub and sub on the same broker (worst case)
    • (75% locality)
    • 50% locality
    • (25% locality)
    • 0% locality: no sub on the broker where we pub
    • equally distributed: pub on the same, subscriber equally distributed on the cluster
  4. Add latency (same on all the links):
    • 0 ms
    • 25 ms
    • 50 ms
  5. Send messages:
    • 100 message per client
    • payload with the timestamp (bytes?)

Topology

  • Fully meshed: done
  • Star: docker swarm/overlay network of docker (???)
  • Tree...

Open points

  • QoS
  • Number of different topics
  • Retained messages