Managing today’s computer networks is a complex and error-prone task. These networks consist of a wide variety of devices, from routers and switches, to firewalls, network-address translators, load balancers, and intrusion-detection systems. Network administrators must express policies through tedious box-by-box configuration, while grappling with a multitude of protocols and baroque, vendor-specific interfaces. In contrast, Software-Defined Networking (SDN) is redefining the way we manage networks. In SDN, a controller application uses a standard, open messaging interface like OpenFlow [1], to specify how network elements or switches should handle incoming packets. Programmers develop their own new controller applications on top of a controller platform which provides a programming API built on top of OpenFlow. Separating the controller platform and applications from the network elements allows anyone—not just the equipment vendors—to program new network control software. In just a few years, SDN has enabled a wealth of innovation, including prominent commercial successes like Nicira’s network virtualization platform and Google’s wide-area traffic-engineering system. Most of the major switch vendors support the OpenFlow API, and many large information-technology companies are involved in SDN consortia like the Open Networking Foundation and the Open Daylight initiative. SDN is creating exciting new opportunities for networksavvy software developers and software-savvy network practitioners alike. But how should programmers write these controller applications? The first generation of SDN controller platforms offer programmers a low-level API closely resembling the interface to the switches. This forces programmers to program in “assembly language,” by manipulating bit patterns in packets and carefully managing the shared rule-table space. In the Frenetic project [2], we are designing simple, reusable, high-level abstractions for programming SDNs, and efficient runtime systems that automatically generate the low-level rules on switches [3, 4, 5, 6, 7]. Our abstractions cover the main facets of managing Figure 1: Software Defined Network (SDN)
[1]
Nick McKeown,et al.
OpenFlow: enabling innovation in campus networks
,
2008,
CCRV.
[2]
David Walker,et al.
Frenetic: a network programming language
,
2011,
ICFP.
[3]
David Walker,et al.
Languages for software-defined networks
,
2013,
IEEE Communications Magazine.
[4]
David Walker,et al.
A compiler and run-time system for network programming languages
,
2012,
POPL '12.
[5]
David Walker,et al.
Abstractions for network update
,
2012,
SIGCOMM '12.
[6]
David Walker,et al.
Composing Software Defined Networks
,
2013,
NSDI.