Issues in wireless data management
Wireless connectivity is fast becoming a reality. Typical mobile and pervasive devices will be able to connect to, and accept connections from, the infrastructure whenever necessary. However, mobile data management has so far focused on disconnected operations, which involves replication of data and batch reconciliation of updates upon reconnection. Emerging wireless pervasive devices enable a new data management model for applications:
- Applications can propagate updates in a trickle fashion, providing several benefits:
- The user experience is improved by reducing the latency of batch synchronization.
- Scalability is enhanced by distributing the workload from clients uniformly over a larger period of time
- Data diverges less, so there are fewer conflicts.
- Dynamic footprint can be reduced because less data has to be buffered at once.
- The infrastructure can push data to applications, with the following benefits:
- The user need not explicitly request synchronization.
- Notification of events of interest to the user is available immediately.
However, there are a number of technical challenges that have to be overcome in order to implement this model:
- Wireless pervasive devices are "connectable" but, unlike desktops, not "connected". There are costs associated with a connection, including setup time, airtime or packet charges, and battery power. Among the questions we want to answer are:
- How can trickle synchronization algorithms be optimized to minimize each of these cost factors?
- How can selective synchronization, propagating only the "most important" updates, be implemented? How should applications specify the selection criteria?
- Which policies and parameters should be under application and/or end-user control?
- How does the system react to variations in network quality?
- What novel approaches are there to make the system robust against communication failures and to allow incremental progress to be made despite failures? What are the costs associated with these and more traditional approaches?
- Pervasive devices are small and severely memory-constrained. Are there space-sensitive trickle-synchronization algorithms that do not overly compromise usability and latency?
[Back to Pervasive Synchronization and Messaging home page.]