Publication

In recent years, we have seen a shift in the way information is processed. Departing from the traditional paradigm in which information is first stored and then queried, we are quickly moving to a new paradigm in which new information is directly routed to the relevant recipients. This new paradigm is being adopted by several research communities, databases being only one of them. Information filters represent one of the key components of this new paradigm, as they loosely couple senders and receivers of data items. Receivers of information submit a profile of their interest to the information filter, while producers of information send messages to the information filter. The purpose of an information filter is then to match the messages to the profiles, so that the matching messages can be sent to the relevant receivers. Information filters are used in areas like application integration, personalized content delivery, networking monitoring and many other areas. Simpler versions of information filters are appearing as products on the market place, while research continues into several directions. In order to enable the information filtering paradigm, techniques like profile indexing and stream processing are used. The main directions of research have been expressiveness of profiles, scalability in terms of profiles and distribution of information filters over networks. This thesis contributes three new aspects to the area of information filtering: scalability in terms of message throughput, context-aware information filters that use state for the matching decision and a study of quality of service. Scalability in terms of message throughput is achieved by processing messages in batches instead of processing them one by one, thus reducing the cost of processing an individual message. Context-aware information filters augment existing, stateless information filters by including context state into the matching decision. Since this state receives updates, the key challenge in building such a context-aware information filter is to deal with high message rates and high update rates. The thesis addresses this challenge in two different ways: AGILE, a method to automatically adapt index accuracy to the workload parameters, and batched processing of updates, where a set of updates is processed at once in order to reduce the cost. Quality of service becomes more and more important as information filters are used in settings where the load is unpredictable and might exceed the available resources. This work examines how state of the art approaches to implement quality of service apply to information filters. For the three areas contributed by this thesis, a theoretical analysis and an extensive performance study is provided, illustrating the benefits and trade-offs. To sum up, this thesis contributes work to improve information filters by increasing the message throughput, including context state in the matching process and studying quality of service. The results provide further support for the adoption of information filters into the mainstream of information processing.

@phdthesis{abc,
	abstract = {In recent years, we have seen a shift in the way information is processed. Departing from the traditional paradigm in which information is first stored and then queried, we are quickly moving to a new paradigm in which new information is directly routed to the relevant recipients. This new paradigm is being adopted by several research communities, databases being only one of them. Information filters represent one of the key components of this new paradigm, as they loosely couple senders and receivers of data items. Receivers of information submit a profile of their interest to the information filter, while producers of information send messages to the information filter. The purpose of an information filter is then to match the messages to the profiles, so that the matching messages can be sent to the relevant receivers. Information filters are used in areas like application integration, personalized content delivery, networking monitoring and many other areas. Simpler versions of information filters are appearing as products on the market place, while research continues into several directions.
In order to enable the information filtering paradigm, techniques like profile indexing and stream processing are used. The main directions of research have been expressiveness of profiles, scalability in terms of profiles and distribution of information filters over networks.
This thesis contributes three new aspects to the area of information filtering: scalability in terms of message throughput, context-aware information filters that use state for the
matching decision and a study of quality of service. Scalability in terms of message throughput is achieved by processing messages in batches instead of processing them one by one, thus reducing the cost of processing an individual message. Context-aware information filters augment existing, stateless information filters by including context state into the matching decision. Since this state receives updates, the key challenge in building such a context-aware information filter is to deal with high message rates and high update rates. The thesis addresses this challenge in two different ways: AGILE, a method to automatically adapt index accuracy to the workload parameters, and batched processing of updates, where a set of updates is processed at once in order to reduce the cost. Quality of service becomes more and more important as information filters are used in settings where the load is unpredictable and might exceed the available resources. This work examines how state of the art approaches to implement quality of service apply to information filters. For the three areas contributed by this thesis, a theoretical analysis and an extensive performance study is provided, illustrating the benefits and trade-offs.
To sum up, this thesis contributes work to improve information filters by increasing the message throughput, including context state in the matching process and studying quality of service. The results provide further support for the adoption of information filters into the mainstream of information processing.},
	author = {Peter M. Fischer},
	school = {16671},
	title = {Adaptive Optimization Techniques for Context-Aware Information Filters},
	year = {2006}
}