Using publish/subscribe middleware for mobile systems

Gianpaolo Cugola, Hans-Arno JacobsenACM SIGMOBILE Mob. Comput. Commun. Rev 2002 · DOI

TL;DR

This paper argues that the publish/subscribe (pub/sub) paradigm is a superior middleware choice for mobile computing applications compared to traditional client/server models. It highlights how pub/sub's inherent decoupling and asynchronous nature effectively address the challenges of mobility, scalability, and dynamic network topologies in modern mobile systems.

What Problem Does This Paper Address?

The paper tackles the growing complexity of mobile computing applications, such as location-based services, sensor networks, and ad hoc networks. These applications require middleware that can handle:

  • Anonymous Entities: Interactions between devices where their identities are not known beforehand.
  • Timely Decoupled Processing: Information needs to be delivered promptly without requiring direct, continuous connections between senders and receivers.
  • Scalability: The ability to mediate interactions between potentially millions of mobile clients.
  • Dynamic Network Topologies: Mobile environments often involve devices joining and leaving networks unpredictably, and network infrastructure itself can be unstable (e.g., ad hoc networks).

Traditional client/server (request/response) middleware struggles with these requirements. A "pull-based" approach leads to constant polling, server overload, and high energy consumption on mobile devices. It also fails to efficiently handle the dynamic nature of information sources.

Methods

The paper advocates for the publish/subscribe (pub/sub) paradigm as the solution. In this model:

  • Publishers send out event notifications (publications) without knowing who will receive them.
  • Subscribers express their interests by issuing subscriptions, which are essentially queries for specific types of events.
  • An event dispatcher (or broker) acts as an intermediary, collecting subscriptions and routing publications to interested subscribers.

The authors differentiate between:

  • Subject-based systems: Subscriptions are based on predefined topics or channels.
  • Content-based systems: Subscriptions are expressions that match the actual content of events, offering greater flexibility and expressiveness. The paper focuses on content-based systems as they are better suited for filtering vast amounts of data for millions of users.

They also discuss two architectural approaches for the event dispatcher:

  • Centralized: A single dispatcher, simpler but a single point of failure and potential bottleneck.
  • Distributed: Multiple dispatchers cooperate to improve scalability and fault tolerance.

The paper then introduces two research prototypes that embody these principles:

  • TOPSS (Toronto Publish/Subscribe System): Focuses on a high-throughput, centralized dispatcher designed to handle millions of subscriptions and high event rates, particularly for event-driven applications. It also explores approximate matching for location-based services.
  • JEDI (Java Event-based Distributed Infrastructure): Addresses scalability and mobility by implementing a distributed dispatcher. It natively supports component disconnection and reconnection, crucial for mobile users. JEDI also explores dynamic adaptation of its routing topology to handle network changes and mobility of dispatching servers.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Pub/Sub as a Natural Fit for Mobile: The core insight is that the inherent decoupling and asynchronous nature of pub/sub middleware perfectly align with the needs of mobile computing, where devices are often disconnected, their locations change, and their interactions are not always pre-planned.
  • Content-Based Filtering is Crucial: For systems dealing with millions of users and events, simple subject-based filtering is insufficient. Content-based subscriptions are essential for precise and efficient information delivery.
  • Mobility Requires More Than Just Disconnection Support: While JEDI's support for moveOut and moveIn is valuable, the paper hints that truly dynamic mobile environments might require more sophisticated adaptations, including the mobility of the middleware components themselves (dispatchers).
  • Approximate Matching for Real-World Data: The introduction of approximate matching in TOPSS (A-TOPSS) is a key idea for handling the inherent inaccuracies and variability of data in mobile environments, especially for location-based services.
  • Scalability Demands Distribution: The paper clearly shows that to handle millions of users and subscriptions, a distributed dispatcher architecture is necessary, moving beyond single-point solutions.

Key Results

The paper doesn't present specific quantitative results from experiments in this overview. Instead, it focuses on:

  • Conceptual Argument: Demonstrating how the publish/subscribe paradigm addresses the identified requirements of mobile computing.
  • Prototype Capabilities: Describing the features and design goals of TOPSS (high throughput, millions of subscriptions, approximate matching) and JEDI (distributed architecture, mobility support, dynamic topology adaptation).
  • Problem Identification: Clearly outlining the challenges in mobile middleware and how pub/sub, through these prototypes, offers solutions.

Limitations and Future Work

The authors acknowledge several limitations and suggest future research directions:

  • Scalability to Millions: While prototypes aim for this, achieving true scalability to millions of users and subscriptions in real-world, highly dynamic mobile environments remains a significant challenge.
  • Server Mobility: JEDI's focus is on client mobility. The paper notes that managing the mobility of the dispatching servers themselves is crucial for highly dynamic ad hoc networks.
  • Dynamic Topology Adaptation: While JEDI has made progress in adapting its tree structure, more robust and dynamic adaptation of routing strategies to changing network conditions and workloads is needed.
  • Diverse Content Formats: The paper mentions the need to support various data formats (HTML, XML, etc.) and heterogeneous notification channels (email, WAP, SMS), which are ongoing challenges.
  • Accounting and Security: Functions like publisher/subscriber accounting and robust security mechanisms (authentication, secure content distribution, confidentiality of events) are identified as open research issues. The confidentiality of events is particularly tricky, as dispatchers need to inspect content for routing.

Summary updated

BibTeX
@article{cugola2002using,
  author = {Gianpaolo Cugola and Hans-Arno Jacobsen},
  title = {{Using publish/subscribe middleware for mobile systems}},
  journal = {ACM SIGMOBILE Mob. Comput. Commun. Rev.},
  year = {2002},
  doi = {10.1145/643550.643552},
}
Figure