POMP:Protocol Oblivious SDN Programming with Automatic Multi-Table Pipelining Chunhui He Xinyu Feng School of Computer Science and Technology State Key Laboratory for Novel Software Technology University of Science and Technology of China Nanjing University hchunhuiemail.ustc.edu.cn xyfengenju.edu.cn Abstract-SDN programming has been challenging because 2)to design flow table layout and forwarding pipelines for programmers have to not only implement the control logic,but new generation SDN to achieve compact flow tables also handle low-level details such as the generation of flow tables and efficient forwarding.As we explain in Sec.III, and the communication between the controller and switches.New generation of SDN with protocol oblivious forwarding and multi- different forwarding pipelines may generate flow tables table pipelining introduces even more low-level details to consider with significantly different sizes We propose POMP,the first SDN programming environment 3)to implement parsers for packets with new header fields supporting both protocol oblivious forwarding and automatic to fit in the underlying protocol oblivious forwarding multi-table pipelining.POMP applies the static taint analysis mechanism.For instance,one needs to generate the technique to automatically infer compact and efficient multi-table pipelines from a data-plane agnostic network policy written by (offset,length)tuples required by POF [2]to the programmer.The runtime system tracks the execution of the locate packet fields. network policy,and automatically generates table entries.POMP There have been many languages proposed to simplify also introduces a novel notion of dependent labels in the taint analysis,which,combined with the runtime information of the SDN programming,but none of them address all the prob- network policy,can further reduce the number of table entries. lems above.Languages such as NetKAT [6],NetCore [7] Like P4,POMP supports protocol-oblivious programming by and Maple [5]try to address the first problem,but they providing a network protocol specification language.Parsers of are designed for Openflow and do not support multi-table packets can be automatically generated based on the protocol pipelining and protocol-oblivious forwarding.P4 [1]provides specification.POMP supports two main emerging SDN platforms, POF and P4,therefore network policies written in POMP are a header specification language for protocol independence. portable over any switches supporting POF or P4. Programmers can write a specification of the header format, from which P4 generate parsers automatically.But P4 is I.INTRODUCTION more of a low-level switch configuration language,and is not Software-Defined Networking (SDN)is a network archi- suitable for high-level controller programming. tecture that decouples control and forwarding.Distributed We propose POMP,a high-level programming environ- switches are managed by a logically-centralized controller, ment to simplify SDN programming.Following the idea of which can be implemented by software through SDN pro- Maple [5],POMP provides a set of APIs and a runtime system. gramming. Programmers can use the APIs to describe the network policy OpenFlow is the first SDN standard,but it only supports a by writing an algorithmic sequential program with a high-level predefined fixed set of networking protocols,and allows only a centralized view of the network environment.This high-level fixed single flow table.New generation of SDN platforms [1], network policy is data-plane agnostic.It is the runtime system [2],[3],[4]offer two new flexible features at the data-plane, that tracks the execution of the network policy and generates i.e.multi-table pipelining and protocol-oblivious forwarding. fow table entries automatically.POMP also makes significant The former allows multiple flow tables on switches to form a extensions to the above ideas in Maple to support automatic forwarding pipeline,which can be customized by users.The multi-table pipelining and protocol oblivious programming. latter supports customized packets for new protocols. As far as we know,POMP is the first high-level programming SDN programming has been challenging because the pro- environment that solves all the aforementioned problems.Our grammer has to not only implement the network policy (i.e. work on POMP makes the following new contributions: the control logic),but also handle low-level data-plane details. Specifically,one faces the following challenges: We apply the static taint analysis technique [8]to anal- yse the network policy's dependence over the fields of 1)to manually translate high-level network policies to flow table entries.It is error-prone,and makes network packets.Based on the fine-grained dependence relation we automatically generate compact and efficient multi- policies hard to write and read [5]. table pipelines.The analysis and the pipeline generation Corresponding author:Xinyu Feng.This work is supported in part by grants from National Natural Science Foundation of China(NSFC)under Grant Nos. IThe name POMP highlights the two key features of our work,i.e.protocol- 61379039and61632005. oblivious programming and multi-table pipelining.POMP: Protocol Oblivious SDN Programming with Automatic Multi-Table Pipelining Chunhui He School of Computer Science and Technology University of Science and Technology of China hchunhui@mail.ustc.edu.cn Xinyu Feng State Key Laboratory for Novel Software Technology Nanjing University xyfeng@nju.edu.cn Abstract—SDN programming has been challenging because programmers have to not only implement the control logic, but also handle low-level details such as the generation of flow tables and the communication between the controller and switches. New generation of SDN with protocol oblivious forwarding and multi￾table pipelining introduces even more low-level details to consider. We propose POMP, the first SDN programming environment supporting both protocol oblivious forwarding and automatic multi-table pipelining. POMP applies the static taint analysis technique to automatically infer compact and efficient multi-table pipelines from a data-plane agnostic network policy written by the programmer. The runtime system tracks the execution of the network policy, and automatically generates table entries. POMP also introduces a novel notion of dependent labels in the taint analysis, which, combined with the runtime information of the network policy, can further reduce the number of table entries. Like P4, POMP supports protocol-oblivious programming by providing a network protocol specification language. Parsers of packets can be automatically generated based on the protocol specification. POMP supports two main emerging SDN platforms, POF and P4, therefore network policies written in POMP are portable over any switches supporting POF or P4. I. INTRODUCTION Software-Defined Networking (SDN) is a network archi￾tecture that decouples control and forwarding. Distributed switches are managed by a logically-centralized controller, which can be implemented by software through SDN pro￾gramming. OpenFlow is the first SDN standard, but it only supports a predefined fixed set of networking protocols, and allows only a fixed single flow table. New generation of SDN platforms [1], [2], [3], [4] offer two new flexible features at the data-plane, i.e. multi-table pipelining and protocol-oblivious forwarding. The former allows multiple flow tables on switches to form a forwarding pipeline, which can be customized by users. The latter supports customized packets for new protocols. SDN programming has been challenging because the pro￾grammer has to not only implement the network policy (i.e. the control logic), but also handle low-level data-plane details. Specifically, one faces the following challenges: 1) to manually translate high-level network policies to flow table entries. It is error-prone, and makes network policies hard to write and read [5]. Corresponding author: Xinyu Feng. This work is supported in part by grants from National Natural Science Foundation of China (NSFC) under Grant Nos. 61379039 and 61632005. 2) to design flow table layout and forwarding pipelines for new generation SDN to achieve compact flow tables and efficient forwarding. As we explain in Sec. III, different forwarding pipelines may generate flow tables with significantly different sizes. 3) to implement parsers for packets with new header fields to fit in the underlying protocol oblivious forwarding mechanism. For instance, one needs to generate the (offset, length) tuples required by POF [2] to locate packet fields. There have been many languages proposed to simplify SDN programming, but none of them address all the prob￾lems above. Languages such as NetKAT [6], NetCore [7] and Maple [5] try to address the first problem, but they are designed for Openflow and do not support multi-table pipelining and protocol-oblivious forwarding. P4 [1] provides a header specification language for protocol independence. Programmers can write a specification of the header format, from which P4 generate parsers automatically. But P4 is more of a low-level switch configuration language, and is not suitable for high-level controller programming. We propose POMP, a high-level programming environ￾ment to simplify SDN programming. Following the idea of Maple [5], POMP provides a set of APIs and a runtime system. Programmers can use the APIs to describe the network policy by writing an algorithmic sequential program with a high-level centralized view of the network environment. This high-level network policy is data-plane agnostic. It is the runtime system that tracks the execution of the network policy and generates flow table entries automatically. POMP also makes significant extensions to the above ideas in Maple to support automatic multi-table pipelining and protocol oblivious programming.1 As far as we know, POMP is the first high-level programming environment that solves all the aforementioned problems. Our work on POMP makes the following new contributions: • We apply the static taint analysis technique [8] to anal￾yse the network policy’s dependence over the fields of packets. Based on the fine-grained dependence relation we automatically generate compact and efficient multi￾table pipelines. The analysis and the pipeline generation 1The name POMP highlights the two key features of our work, i.e. protocol￾oblivious programming and multi-table pipelining