Some Future Research Directions SIGMETRICS 2007 Don Towsley

Some Future Research Directions SIGMETRICS 2007 Don Towsley UMass-Amherst

Overview q PE concerned with solving problems v implications? some challenges q education for the system q

PE confluence of many areas PE problem solving nail -> hammer screw -> screw driver nut -> wrench design exploration -> stochastic models measurements -> statistics resource allocation > optimization theory dynamic rsrc alloc -> control theory

optimization machine learning game theory PE statistics stochastic processes control theory signal processing information theory

Information theory and PE IT concerned with minimizing communication resources q entropy – communication usage bound q q sensor networks characterized by v severe resource constraints v highly correlated data streams v network monitoring, radar networks, habitat sensor nets, …

Query processing in data sensor networks Challenge: given set of queries, minimize resource consumption to satisfy query result metric Resources: bandwidth, power, processing, storage Metrics: error in result (rate distortion), power consumption, … Issues: complexity, resource constraints Tools: traditional PE, information theory, control theory, ML, …

PE, control optimization, game theory Many PE problems are optimization problems q storage management q call admission q congestion/flow control Often between competing parties Need to address entire problem – not just evaluate performance of one instance

Multiple controllers q network control v routing, congestion control, call admission add an overlay q and another q Control

Multiple controllers q network control v q q q routing, congestion control, call admission add an overlay and another or an application Result? v controller mismatch? v well-tuned machine? v performance implications? Control

Multiple controllers Issues: complex interactions among selfinterested players Tools: traditional PE, control theory, game theory, economic theory

Training for PE q background in v probability statistics q theory, stochastic processes, course(s) in performance evaluation v how to handle real world problems – right questions? assumptions v iterative modeling/validation process v combining analysis, simulation, measurements use good case studies q exposure to (some of) v ML, information theory, convex optimization, differential equations, game theory, control theory, …

Thanks! Questions?