SNSF Starting Grant awarded to Prof. Katerina Argyraki
© 2015 EPFL
Katerina Argyraki, Professor at the Laboratory Network Architecture, has been awarded a STARTING GRANT by the SNSF in the context of the so-called “Temporary Backup Schemes” for the European Research Council (ERC) grants. This transitional measure of the SNSF offers the awardee an adequate substitute for ERC grants.
Abstract - Reconciling Flexibility and Predictability in Computer Networks
Today's society depends on computer networks more than ever, and the accelerating rate at which new uses of these networks are being designed has surpassed the ability of the networks themselves to keep up -- they start suffering from fundamental architectural limitations. Yet the networks themselves must evolve with the needs of their users and operators, and this is especially true for the Internet, which influences the quality of life of billions of people. The more our society depends on networks for services we consider fundamental human rights (e.g., education, medical care, freedom of speech, private communication), the more important it is that these networks preserve their ability to serve users in the face of unforeseen attacks and technical problems. For example, as new security threats emerge, the functionality of the Internet must evolve to recognize them and protect its users. As adversaries evolve, so must the network fabric: it must provide new functionality without requiring a "forklift upgrade."
Our objective is to fundamentally change the design of network equipment, so as to make it amenable to repeated and continuous reprogramming, thus allowing network functionality to evolve together with the needs of its users and operators, as well as with the capabilities of adversaries. This requires a radical departure from today's architecture that forces a choice between flexibility and performance+predictability. We propose the *virtual data plane*, a way to combine a custom hardware layer for fast packet processing with a programmable software layer. Programmability alone, however, is not enough: a truly flexible network fabric must support programmability while at the same time guaranteeing security, performance, and energy efficiency. Network programs should be verifiably isolated from each other, and the operator of the network equipment should receive guaranteed bounds on the resources consumed by each network program.
We expect our approach to have the same profound impact on programmability of network equipment that virtual memory had in the 1960s on the programmability of general purpose computers: by virtualizing caches, RAM, and disk, and then delegating management of the memory hierarchy to the operating system, it became possible to write programs that interact with just one, simple, direct-addressable memory.
Details of the Grant
Requested Funding: CHF 1’391’816.-
Duration: 60 Months
Host Institution: EPFL
Domains: Engineering Sciences, Information Sciences