mirror of https://github.com/sigp/gossipsub-testground.git synced 2026-01-08 05:03:58 -05:00

Files

Age Manning a260f69580 Gossipsub Simulations (#9 )

* episub-sim

* network generation

* mainnet network generation

* no pretty print

* add small network for testing

* upgrade versions, make the thing compile

* checkpoint

* this is ridiculous

* make gen_topology reachable by testground

* add 16 instsances for testing

* add running command

* remove params

* pass new participants param

* update upstream reg

* update upstream reg

* put msg generation back in

* put publishing back in

* adding metrics

* record metrics on intervals

* add gossip max limit

* stop fghting testground. For now

* update README

* Revert "stop fghting testground. For now"

This reverts commit a976c5371b.

* use composition files to get the docker build context a layer up

* fix params and logs

* remove unused files

* cache workspace deps

* cache workspace deps _the right_ way

* updates

* some docs

* Update CI and remove root workspace

* Add duplicates, fix clippy, improve dash

* Reduce message sizes, some debugging

* Update dash and logs

* Add scripts folder

* Fix executor lockup

* fmt and clippy

* Add some docs

* Dot to mermaid

Co-authored-by: Diva M <divma@protonmail.com>

2022-11-29 16:06:53 +11:00

2.8 KiB

Raw Permalink Blame History

Censoring Simulation

This simulation creates a number of malicious nodes which do not propagate received messages. This simulation is aimed to help fine-tune gossipsub scoring parameters to mitigate censoring attacks on gossipsub networks.

Running the simulation

testground run composition -f censoring/compositions/composition.toml --wait

How the Simulation Works

Note: Attackers connect to a single publisher (victim). Publisher1 is the victim in this test plan.

sequenceDiagram
    participant Publishers
    participant Lurkers
    participant Attackers

    %% Discovery
    Note over Publishers,Lurkers: Setup discovery<br />Connect to some of the honest nodes (publishers + lukers)<br />randomly selected
    Note over Attackers: Setup discovery<br />Connect to a single publisher
    Publishers->>Lurkers: Connect
    Lurkers->>Publishers: Connect
    Attackers->>Publishers: Connect to a single publisher (victim)

    %% Subscribe topics
    Note over Publishers,Lurkers: Subscribe topic(s)
    Publishers->>Lurkers: Subscribe/GRAFT
    Lurkers->>Publishers: Subscribe/GRAFT
    Publishers->>Attackers: Subscribe/GRAFT
    Note over Attackers: Subscribe to the topic in the message from the publisher, <br/>and send back Subscribe/GRAFT.
    Attackers->>Publishers: Subscribe/GRAFT

    %% Publish messages
    Note over Publishers: Periodically publish messages on all topics subscribing.
    loop Publish messages
        Publishers->>Lurkers: Message
        Publishers->>Attackers: Message
        Note over Attackers: **Don't propagate messages**
    end

    %% Record metrics
    Note over Publishers,Lurkers: Record metrics.

Dashboards

Please see the root README for how to run Grafana.

Gossipsub Metrics

The metrics of gossipsub are recorded once the simulation has been completed. All of the metrics on libp2p-gossipsub are available in this dashboard.

Variables for this dashboard:

run_id: The run_id for the test run you want to see.
topic: The gossipsub topic, currently it's fixed to emulate.
instance_name: Some panels in this dashboard need an instance name to be specified. (e.g. score_per_mesh histogram)

Peer Scores

The peer scores are recorded periodically (every second) while the simulation is running.