Exploring online evolution of network stacks

Imai, Pierre. Exploring online evolution of network stacks. 2013, Doctoral Thesis, University of Basel, Faculty of Science.


Official URL: http://edoc.unibas.ch/diss/DissB_10689

Downloads: Statistics Overview


Network stacks today follow a one-size-fits-all philosophy. They are mostly kept unmodified due to often prohibitive costs of engineering, deploying and administrating customisation of the networking software, with the Internet stack architecture still largely being based on designs and assumptions made for the ARPANET 40 years ago. We venture that heterogeneous and rapidly changing networks of the future require, in order to be successful, run-time self-adaptation mechanisms at different time scales and based on continuous performance measurements.
For this purpose we present an autonomous stack composition framework and configuration logic inspired by biological evolution: Stack configurations (compositions) compete against each other on-line, new compositions evolve from the best previous performers. Compositions are further selected and pooled together according to the traffic and network conditions, forming a long-term situation-aware knowledge base.
We demonstrate the feasibility of our runtime adaptive approach by exposing our implementation to simulated as well as real world Internet traffic. Beyond individual “zero knob protocols” we show that our network management system not only tunes a network stack’s parameters but can also change its composition on the fly.
This lowers the barrier for introducing novel protocols, move to other run-time systems or accommodate new traffic patterns. Ultimately, this lets engineers of future computer networks focus on specialised rather than smallest common denominator solutions, as the run-time choice and management is taken care of by our system.
Advisors:Tschudin, Christian
Committee Members:Plagemann, Thomas
Faculties and Departments:05 Faculty of Science > Departement Mathematik und Informatik > Informatik > Computernetzwerke (Tschudin)
UniBasel Contributors:Imai, Pierre
Item Type:Thesis
Thesis Subtype:Doctoral Thesis
Thesis no:10689
Thesis status:Complete
Number of Pages:244 S.
Identification Number:
edoc DOI:
Last Modified:22 Jan 2018 15:51
Deposited On:31 Mar 2014 09:59

Repository Staff Only: item control page