SYNC: Synergistic Network Policy Management for Cloud Data Centres

点击翻译按钮获取中文摘要

All computer networks, including cloud data centre networks, are governed by high-level policies derived from network-wide requirements, such as "file servers should only be accessible by internal IP (Internet Protocol) addresses". Upon deployment, an individual policy is realised as a composition (or chain) of network packets processing rules that will be placed in a specified sequence of network function boxes within the network.Traditionally, implementing network policy is an error-prone manual configuration process. Emerging technologies such as Software-Defined Networking (SDN) and Network Function Virtualisation (NFV) have largely eliminated the need of manual configuration through software automation. Nevertheless, the use of SDN and NFV have resulted in greater number of independent network nodes that dynamically generate and implement policy rules respectively, making correct policy implementation a hard problem to solve. Worse still, this problem is amplified by the dynamic virtual machine (VM) consolidation in cloud data centre since migrating VMs means that the "specified sequence" must also be updated across the network. Imperfect policy implementation will lead to policy violation that attributes to 78% of data centre downtime, which costs $5,600(£3,758) per minute. This demonstrates the necessity of synergistic placement of network policy rules and application VMs, as captured in the SYNC project hypothesis: Infrastructural configuration and utilisation as well as application performance of cloud data centres is largely dictated by the placement of network policy rules and application virtual machines. This is in contrast to the existing approaches which either only consider static rules placement, or perform dynamic placement without taking into account the application VMs.In this project, we propose the development of SYNC, a synergistic network policy management framework that will lever synergy amongst a) policy rules, b) applications and c) underlying temporal network state for achieving network-wide performance optimisation. In order to realise SYNC, the following research and development tasks will take place: i). The high-level network policy expressions will be decomposed into minimum set of network-wide consistent chains of rules, which are in turn implemented in network function boxes at different network locations, e.g. middleboxes, network switches, and end hosts. ii). The underlying network state will be exploited to (re)arrange application virtual machines and rules so that the network-wide impact of pairwise traffic patterns is minimised. The key challenge in this innovation will be the scale of the underlying infrastructure, which can have up to a million VMs and millions of rules. We will construct appropriate models and efficient algorithms, combined with SDN and NFV overcome this challenge. We intend to publish research outcome in prestigious journals and conferences, provide open-access to the research data, and commercialise our intellectual property.