Many distributed applications are characterized by having several concurrent data sources, each data source sending a stream of messages to the same large set of receivers. This is a typical deployment pattern in consistent replication schemes, distributed data stream processing, peer-to-peer bulk data transfer, etc.
When multiple data sources partition away at once, the failover mechanism includes synchronization of the received data across all the receivers. A typical solution is that a backup server takes over the partitioned data source and retransmits all of the missing messages. However, this solution was discovered to create a bottleneck in industrial large-scale deployment scenarios, such as those of the WebSphere Application Server.
If the receivers are able to cooperate and share the burden of retransmissions, a much more efficient solution can be devised. The objective of this project is to complete the design of a cooperative synchronization scheme that balances the load of retransmissions, as well as implement it and verify through evaluation.
The work can be divided into two tasks:
- Implementation of an algorithm that computes the optimal job assignment for the retransmission task. There will be need to implement several versions of the algorithm and compare them through evaluation.
- Integrating the algorithm into existing applications or middleware. Perhaps the most straightforward application choice would be to incorporate the algorithm into an existing group communication toolkit but other applications are also possible.
Program
Informatikk