CAREER: An Architecture for Building Dynamic, Adaptive Systems

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

This is an NSF CAREER/PECASE grant proposal for research into a new Architecture for wide-area, Dynamic Adaptive Systems (ADAS). The decreasing cost of computing and networking technology and network-enabled devices is enabling the large-scale deployment of a wide-area infrastructure that operas many new and innovative services (e.g., a service that allows a user to control the lights in a room). Unfortunately, wide-area network deployment has not yielded the expected improvements in access to new and existing devices, services, and networks. Each of these areas faces significant problems: end devices that are more powerful than earlier generations, but are still very limited and fragile relative to their fixed counterparts; many services are developed with the static assumption of high-bandwidth and low-latency networks and thus perform poorly in the low bandwidth, high-latency wide-area environment; networks that have different capabilities (e.g., bandwidth, latency, cost, coverage) making reconciliation of these differences difficult for many services. This CAREER grant proposal addresses two key problems of wide-area services: (1) Dynamically adapting to the current conditions and environment at each level of the service and network, and (2) Locating the appropriate service for a task. Adaptive systems are a necessary response to the significant performance problems associated with accessing services in the wide-area. A dynamically adaptive system continually changes based upon changing conditions at multiple levels: link, network, and service, yielding a significant improvement in the perfor- mance and usability of services. Most networks make a fixed tradeoff between reasonable performance under worst-case conditions and performance under best-case conditions. However, for a fixed design point, best-case performance usually suffers. Previous research efforts have primarily focused on only one aspect of adaptation. What makes this grant proposal unique is a plan for developing a new architecture that supports novel, dynamically adaptive services and enables anywhere, anytime, any mode access to services. The first step is to produce software tools, applied theoretical and simulation results, and an experimental testbed, all of which will aid researchers working with new models for wide-area service construction, deployment, and access. The second step is to use more mature versions of these tools in a large-scale experiment focused on two areas: improving the undergraduate and graduate academic experience in two courses and providing new undergraduate research opportunities in the intersection between mobile computing, wireless networking, and wireless telephony. This work builds upon the our research in the following areas: (1) Information Exposure and Multi-level Dynamic Adaptation. A key enabler for adaptation is the exposure of metadata between link, network, and service levels (e.g., a service can inform the link layer of its latency and reliability quirements). This technique will depend upon the development of theoretical models for the predicted state of a wireless link, upon based upon extensive experimental and simulation research. (2) Providing high-performance, wide-area access to services. The researcher strongly believes that "Access is the killer application." Decoupling user interfaces from communication and dynamic adaptation are the keys to delivering high-performance. (3) Large-Scale Deployment of Wide-Area Information Access. An important component of the research is conducting large-scale testbed experiments in both research and academic settings. Graduate and undergraduate students will be included in the project's research by leveraging the significant hardware and software testbed being deployed by related projects at UC Berkeley. These research directions will work together in a synergistic fashion, where theoretical, analytical, and simulation results guide the development of useful software infrastructure components | components that will be deployed in a large-scale research and academic testbed, providing a unique opportunity for significant research and academic use of the results by both undergraduate and graduate students. Since the ADAS framework is really an abstraction and synthesis of many earlier ideas, the research results from this effort should be of interest to users of a broad class of wide-area information access services and applications, not merely those who choose to use ADAS software. These results will be applied within the ADAS architecture to a variety of service and information access challenges.