I consider research as fundamentally similar to learning. Both are driven by curiosity and a desire for understanding the course of the world. As we grow up through years of parental teaching and formal schooling, the topics of our study become gradually more complex and specialized. The type of learning gradually changes from "monkey see monkey do", to structured knowledge transfer, to self-directed exploration.
This process culminates in what should be academic research, where we try to learn and discover new ways of understanding, describing, operating, and changing the matters of the world, and share newly found knowledge. In contrast to conventional learning, the challenges studied are inherently novel, so there is nobody to ask for an existing solution. Further, knowledge is less certain, so it needs factual proof and corroboration that it is useful. From that perspective the duality of teaching and research becomes a single spectrum of knowledge discovery, preservation, and transfer.
My primary research interests are network and distributed systems architecture, protocols, services, technology, and implementation. In terms of meta-topics, I find a detailed and well-substantiated improvement in understanding or addressing a well-established problem at least equally intriguing as formulating an outlandish problem (or redressing a well-known one) to claim a novel approach.
Much has been said and written about the potential shortcomings of the current Internet architecture, so I don't want to add more here. It is clear that any large-scale system architecture needs to be continuously re-evaluated to find possible improvements.
To this end, my main interest is in finding a basic formalism (a language of sorts) to precisely describe essential network functionality, such as forwarding, routing, and naming/discovery. Given that even basic concepts such as name and address lack a precise and commonly agreed definition, it is high time that networking is put on a theoretically sound footing. Ideally, such a language would also permit code generation and verification and thereby ease the deployment of new communication protocols and services.
Recent Publications
The quality of the network service determines what end-to-end transport services can be offered to applications. It is not clear what quality of network service is necessary to support the potentially large variety of applications and communication scenarios now and in the future. Valid proposals range from extremely simple (best-effort and over-provisioning) to very complicated (Integrated Services and detailed accounting). In general, the building blocks for network QoS are: admission control and traffic regulation, packet scheduling, and traffic engineering through constraint-based routing. In past work, I have studied edge-based load control, fine-grained scheduling with low overhead, and innovative packet scheduling approaches. I keep an interest in the area, in particular buffer management and feedback-based load control.
Recent Publications
Implementation details and performance of network technology usually boils down to two issues: buffer management and asynchronous execution. In the past, I have studied two specific cases of networking mechanisms and devised schemes for high-performance execution. In the future, I would like to generalize some of the fundamental findings and lessons learned. For example, I am now keenly interested in computer runtime environments in general, especially in a distributed environment. This term is meant to encompass traditional operating systems, middleware, and virtualized systems. The goal of this research is the systematic study of runtime environment across a spectrum ranging from multi-core to classical distributed system. How can these systems be designed and programmed, such that a runtime environment has good information to perform resource allocation decisions? This also includes performance modellung and measurement.
Recent Publications
last updated Feb 19, 2010 - Martin Karsten