FuseExchangeProtocol.jl

FUSE eXchange Protocol defines the handshake for processes interfacing with FUSE, for example for co-simulation purposes with a controller.

sequenceDiagram
    autonumber
    actor FUSE as Service Requestor
    participant REDIS
    actor TokSys as Service Provider

    rect rgb(255, 223, 191)
        TokSys--)REDIS: Register service X (subscribe to channel X)
    end
    
    rect rgb(223, 255, 191)
        FUSE-->REDIS: Is service X available? (pubsub channels)
    end

    rect rgb(255, 191, 223)
        FUSE->>REDIS: del ID__X__req2pro
        FUSE->>REDIS: del ID__X__pro2req
        FUSE--)TokSys: publish FUSE session ID to channel X
    end
    
    loop
        rect rgb(191, 223, 255)
            FUSE->>+REDIS: lpush service inputs
            note over REDIS: ID__X__req2pro
            REDIS->>-TokSys: brpop service inputs
        end
        rect rgb(191, 223, 255)
            TokSys->>+REDIS: lpush service outputs
            note over REDIS: ID__X__pro2req
            REDIS->>-FUSE: brpop service outputs
        end
    end

Loading

FuseExchangeProtocol builds on top of REDIS

• High Performance (operates in-memory) <1ms latency when run locally
• Support large data volumes (<512 MB per message)
• Distributed (ie. not geo-located)
• Versatility of data structures (lists, queues, streams, timeseries,...) with atomic operations
• Pub/Sub System
• Synchronous / Asynchronous communication patterns
• Supports multiple producers / consumers pattern
• Horizontal and vertical scalability
• Multi-language client libraries and open protocol
• Used by Twitter, GitHub, Snapchat, Airbnb, Netflix
• In comparison to other tools:
  • Unlike Memcached, REDIS supports a wider range of data structures and persistence.
  • While Kafka and RabbitMQ are more focused on message queuing and streaming, REDIS offers these capabilities along with its role as a data store and cache.
  • Compared to database services like DynamoDB, REDIS can serve as a more immediate, low-latency layer for data access and manipulation.
  • Unlike Hazelcast and etcd, which are more focused on distributed computing and configuration management respectively, REDIS offers a more general-purpose approach with its data structure support and performance.

Online documentation

For more details, see the online documentation.