Room: Mackenzie Building, Phone: 520-2600 x1749
Email : greg@sce.carleton.ca
URL: http://www.sce.carleton.ca/faculty/franks.html
No./Title: FRANKS-1. Trace-based Performance Model Generation.
Level: Ph.D. and Master's
Background: SYSC-5101
Supervisor: Professor R.G. Franks
Synopsis: This project looks at the problem of constructing performance models of existing distributed and concurrent software systems. Models of these systems are often useful for both capacity planning and system redesign. However, significant effort is often required to find the interactions in the system and the parameters necessary to build a performance model. The goal of this research is to discover ways to automate this procedure so that existing systems can be modelled. Three major steps are required. First, a system must be instrumented so that interesting performance events can be extracted. The event monitoring system should be as unobtrusive as possible to minimize its affect on the system under test and to facilitate its use by users in the field. Second, the events must be translated into a performance model. The resulting model will likely have to undergo transformation in order to simplify it so that it can be analyzed parametrically. Finally, an expert system will have to be developed to interpret the results of the automatically generated model, and perhaps ofter hints at improving the target system's performance.
No./Title: FRANKS-2. Component-based Performance Analysis for Complex Internet Telephony
Level: Ph.D. and Master's
Background: SYSC-5101
Supervisor: Professor R.G. Franks, Professor C.M. Woodside and Professor D.C. Petriu
Synopsis: Internet telephony systems will be engineered out of distributed software components, partly configured statically and partly configured on the fly as a call is handled. Performance questions such as delay and capacity will be much harder to answer, than for systems which are engineered as a single product. This project will use layered performance models for systems with many possible configurations, and consider how to deal with the resulting complexity in practice. It will develop methods for simplifying the analysis by clustering the configurations, and by hierarchical organization of submodels. It will apply these methods to emerging architectures involving SIP, web services, middleware, and distributed feature composition. These projects are funded by CITO and are done in close cooperation with AT&T.
No./Title: FRANKS-3. Performance Analysis of Peer-to-Peer Applications and Protocols
Background: SYSC-5101
Supervisor: Professor R.G. Franks
Synopsis: This research considers distributed systems, in particular those exhibiting peer-to-peer interactions such as internet routing and peer-to-peer protocols. The research will develop new algorithms within the layered queueing network paradigm, to study these types of interactions. Further, this research will apply the new algorithms to the domain of routing and file sharing protocols to gain insights into the behaviour of these protocols to predict their performance and scalability.
No./Title: FRANKS-4. System Optimization using Layered Queueing Models
Level: Ph.D. and Master's
Supervisor: Professor R.G. Franks
Synopsis: This project looks at studying optimization of a system through the solution of multiple performance models. The goal is to use the solution of layered queueing models as the objective function to an optimization technique such as genetic algorithms, simulated annealing or branch and bound.
No./Title: FRANKS-5. Solution of Analytic Performance Models
Level: Ph.D. and Master's
Background: SYSC-5101
Supervisor: Professor R.G. Franks
Synopsis: This project looks at the algorithms for solving performance models analytically. Fast and accurate solutions to performance models are necessary to allow performance experts to model systems. The goal with this reasearch is to broaden the capabilities of, and increase the accuracy of, analtytic solutions. Some of these areas are networking delays, simultaneous resource possession and thread interactions.