MediaBroker: An Architecture for Pervasive Computing

1
MediaBrokerAn Architecture for Pervasive
Computing
Kishore Ramachandran (http//www.cc.gatech.edu/ra
ma) College of Computing Georgia Tech Presented
by Pouria Pirzadeh and Ronen Vaisenberg
2
Computing/Communication Continuum
Sensor Network
HPC resources
Ambient Computing Infrastructure
3
Application Characteristics

• Physically distributed heterogeneous devices
• Interfacing and integrating with the physical
  environment
• Diverse stream types
• Diverse computation, communication and power
  capabilities (from embedded sensors to clusters)
• Stream fusion/transformation, with loadable code
• Resource scarcities
• Dynamic join/leave of application components

4
MediaBroker

• An architecture for stream management
• A clearing house for sensors and actuators in a
  given space
• Stream registry, discovery, sharing
• Dynamic connection of sources (producers) and
  sinks (consumers)
• Dynamic sharing of transformations and streams
• Architecture Requirements
• Scalability
• Low latency/high throughput
• adaptability

5

• Elements
• Type server stores data types, relationships,
  and transformation code
• Transformation engine allow safe execution of
  injected code on cluster nodes
• Scheduler manages workload, and allows
  prioritizing transformation requests
• Data brokers manages connections between
  producers and consumers

Transformation Engine
Data Broker
Data Broker
Data Items
Transformation Requests
Transformation Code
Transformation Engine
Scheduler
Transformation Engine
6
Architecture Overview
7
Architecture Workflow

• New client connects to engine
• MB establishes a command channel
• Associates a listener to the channel
• Client API
• Sources/sinks instantiation/termination
• New source instantiated
• MB assigns it to a new data broker
• New sink instantiated
• MB assigns it to an existing data broker

8
Data Broker

• Core of MB engine
• An execution context, containing 4 threads
• Transport thread
• From source to sink(s)
• Command thread
• Listens to channel for add/remove source/sink
• Source-watch thread
• Sink-watch thread
• Tasks
• Transports data from one source to sink(s)
• Type negotiation between source/sink(s)

9
Data Types

• Data broker
• finds LUB of requested types
• Communicates LUB with producer
• It does LUB transformation to requests
• Data types defined by a language for
• Type description
• Domain attributes
• Type relations
• Through a Type map
• Type transformations

10
MediaBroker Vs. Satware

• Code
• C vs. Java
• Types
• C struct vs. tuples
• Resource Discovery
• Ldap vs. RMI
• Messeging
• Channel listeners vs. RMI

RMI