Periodic Reporting for period 3 - SyNET (From Network Verification to Synthesis: Breaking New Ground in Network Automation)
Reporting period: 2022-11-01 to 2024-04-30
The project is addressing the general challenge of building provably-correct network infrastructures. This is an important challenge as, each year, billions of Euros are lost due to avoidable, human-induced network downtimes. Worse yet, some of these downtimes affect critical infrastructures such as preventing emergency calls from being placed. The unique approach of this project is to leverage formal network models to automatically synthesize correct and efficient network configurations from high-level requirements. In addition to synthesizing correct configurations, these models will also be used to verify running network infrastructures, thereby allowing to improve existing network deployments. Technically, the project sits at the intersection of several areas of computer sciences including: computer networks, programming languages, and software verification. Regarding concrete objectives, the project aims at developing an end-to-end configuration synthesizer platform which is: (i) optimal (it generates provably-optimal configurations); (ii) easy-to-use (for network operators); and (iii) expressive (it supports a vast collection of network protocols and mechanisms).
Thus far, the project has achieved a number of key milestones already, including:
- Synthesis of configuration updates: We adressed the problem of how to seamlessly reconfigure a running network. We proposed new algorithms (and proved their correctness) and developed a system capable of reconfiguring live networks. This work led to an accepted paper at ACM SIGCOMM, a top-tier venue in the field of computer networks.
- Verification of network models: We adressed the problem of assessing the correctness of the network models we use to synthesize configurations. We developed an end-to-end system and use it to find actual bugs in state-of-network network verifiers. This word lead to an accepted paper at USENIX NSDI, another top-tier venue in the field of computer networks.
- Computational complexity: We analyzed the computational complexity of synthesizing network configurations. While we proved that many synthesis problems are NP-hard, we also identified some instances that can be solved efficiently (in polynomial time), opening up the way for developing efficient synthesizers. This work is currently under submission at a top-tier venue.
- Generalizing network models: We are currently developing the first network models capable of reasoning about transient correctness and performance. Doing so will pave the way to synthesize configurations compliant with functional and non-functional requirements. We are currently working on two full-length submissions in that space, both of which should be submitted by the end of 2022.
- Synthesis of configuration updates: We adressed the problem of how to seamlessly reconfigure a running network. We proposed new algorithms (and proved their correctness) and developed a system capable of reconfiguring live networks. This work led to an accepted paper at ACM SIGCOMM, a top-tier venue in the field of computer networks.
- Verification of network models: We adressed the problem of assessing the correctness of the network models we use to synthesize configurations. We developed an end-to-end system and use it to find actual bugs in state-of-network network verifiers. This word lead to an accepted paper at USENIX NSDI, another top-tier venue in the field of computer networks.
- Computational complexity: We analyzed the computational complexity of synthesizing network configurations. While we proved that many synthesis problems are NP-hard, we also identified some instances that can be solved efficiently (in polynomial time), opening up the way for developing efficient synthesizers. This work is currently under submission at a top-tier venue.
- Generalizing network models: We are currently developing the first network models capable of reasoning about transient correctness and performance. Doing so will pave the way to synthesize configurations compliant with functional and non-functional requirements. We are currently working on two full-length submissions in that space, both of which should be submitted by the end of 2022.
By the end of the project, we are expecting to:
- develop a theory of network-wide configuration synthesis, including which problems are possible to solve in the space, and with what complexity.
- develop general techniques and systems for synthesizing robust and performant network configurations. This has not been done up to now, yet is critical.
- develop the first end-to-end system capable of reasoning about probabilistic network, possibly transient, network behaviors.
- develop a theory of network-wide configuration synthesis, including which problems are possible to solve in the space, and with what complexity.
- develop general techniques and systems for synthesizing robust and performant network configurations. This has not been done up to now, yet is critical.
- develop the first end-to-end system capable of reasoning about probabilistic network, possibly transient, network behaviors.