Page tree

Have questions? Stuck? Please check our FAQ for some common questions and answers.

Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 2 Next »

Brigade Leads:

  • Carmelo Cascone / ON.Lab (carmelo@onlab.us)

Brigade Members:

  • Andrea Campanella / ON.Lab (andrea@onlab.us)
  • Yi Tseng / ON.Lab (yi@onlab.us)
  • Jonghwan Hyun / ON.Lab (...)

Contact the brigade:

For any information or to join the brigade please contact Carmelo Cascone (carmelo@onalb.us) or David Boswell (david@onlab.us).

Background:

P4 is a domain-specific language (DSL) designed to allow the programming of packet forwarding devices. P4 can be used to program different targets such as software switches, FPGA-based NICs or switches based on reconfigurable ASICs. P4 enables protocol-independent programmability at different levels, for example:

  • Parsing and modification (actions) of new, non-standard headers.

  • Configure table properties such as size, type of match (exact, ternary, LPM), counters.

  • Stateful processing, i.e. per-packet custom actions that can access and manipulate state maintained by the switch.

P4 allows programming of many devices in an target-independent manner, using high-level constructs. In principle, P4 programs should be portable. The same program when compiled for different targets should produce the same forwarding behavior. Moreover, P4 allows for reconfigurability in the field. In other words, once deployed, a P4-enabled device can be reconfigured with a new P4 program to provide support for new forwarding capabilities.

Scope:

The scope of this brigade is to make ONOS aware of this new dimension of programmability, in which support for new forwarding/processing capabilities can be added by writing and deploying P4 programs.

The P4 paradigm is very different from the current fixed-function one derived from OpenFlow and abstracted by ONOS. Today, capabilities are tightly coupled to a specific device vendor and model. Different devices expose different capabilities, e.g. match on different headers and support for different actions. Once the network infrastructure has been deployed, capabilities cannot be changed, ONOS can just control them, e.g. installing flow rules. Instead with P4, the capabilities of a device can change in time. P4 programs can be provided by switch vendors, operators, applications or ONOS itself. Moreover, the type of capabilities that can be described with P4 go well beyond that described by OpenFlow.

ONOS APIs and services should capture this new dimension of flexibility. On how this should be done, this is up to the brigade. Before explaining the roadmap, it is useful to clarify two points.

Why ONOS should care about a programming language?

P4 is becoming the common language spoken by switch vendors and operators to agree on what the data plane can or should do. Indeed, P4 is meant as both a specification language, e.g. to formally specify how a fixed-function switch ASIC works, and a programming language. In its mission to ease the life of operators, and to promote faster innovation in the network, ONOS should be able to understand and potentially speak P4. Understand, i.e. to parse P4 programs to be aware of the capabilities of a given device and expose higher-level APIs to control them. Speak, i.e. to generate or modify existing P4 programs that can be later controlled to satisfy the needs of applications.

Runtime control of P4 devices.

P4 is not a protocol or device API for runtime control, i.e. once a P4 program is deployed to a device, P4 doesn’t tell us how that device can be controlled, for example to add or remove entries in match+action tables, or to read the value of a counter. How can ONOS control a P4-enabled device? There is no standard control protocol/API yet specifically targeted for P4, so far it is up to the switch target to expose a suitable API. P4Runtime is an effort in the P4 community to create a standard control-plane API portable across targets, they propose a gRPC-based APIs (p4runtime.proto).

owever, different devices supporting P4 might expose different APIs. Similarly to how ONOS today deals with different flavors of OpenFlow exposed by different device vendors/models, heterogeneity of control protocol/APIs should be abstracted from application

Roadmap

[work in progress]

Short-term focus in 2017:

  •  support for P4Runtime (via BMv2)
  • Define and implement use cases to assess southbound support

How to get involved:

Support for P4, and protocol-independence in general, will affect the whole ONOS platform, fromt the southbound to the northbound. The P4 brigade is looking for members willing to contribute! Contact Carmelo Cascone (carmelo@onalb.us) or David Boswell (david@onlab.us) if you are interested.

  • No labels