Bithika Khargharia shares an overview of three open source projects from ONF’s OpenSourceSDN.org.
With SDN and a disaggregated network stack, interoperability is what enables best-of-breed choice and prevents vendor lock-in when the pieces are aggregated back in the context of use cases and solutions. This week I want to talk about three different ONF OpenSourceSDN.org projects that are driving this interoperability in – programmable data planes (Project Protocol-Independent Forwarding (PIF)), microwave and millimeter wave networks (Project Centennial), and packet-optical networks (Project Snowmass).
PIF brings interoperability to programmable switches, a new generation of switches whose data plane functionality can be configured at compile-time using a protocol-neutral instruction set and a high-level programming language such as P4, PX, PIFL, etc. At runtime then the OpenFlow protocol can be used to control the switch’s processing of packets. This model is a contrast to the traditional fixed-function switches where the OpenFlow protocol needs to support the capabilities of the switches bottom-up as a way to control runtime processing of packets. With the new breed of programmable switches, their desired behavior in the network can be configured top-down, with the capabilities of a runtime control protocol like OpenFlow being configured in tandem, rather than being built into a slowly evolving protocol specification. This is where Project PIF is further pushing the boundaries of interoperability. As shown in the figure, PIF is working on the guts of the compiler that can take data plane configuration programs written in any high-level packet processing language (P4, PX, PIFL, etc.) and generate the target binary for a variety of programmable data plane targets. A frontend takes the program, maps it to an Intermediate Representation (based on an abstract packet processing model) which is then mapped by a backend to the desired binary. This mechanism obviates the need to develop a specialized compiler per programming language and data plane target combination, thereby reducing compiler explosion. This also provides an opportunity for other high-level programming languages to surface in the future and for new programmable targets to easily plug into the SDN ecosystem. This is the added value of the PIF work over and above what is described here.
Project Centennial expands interoperability to yet another domain of networking – microwave and millimeter wave transport networks. It specifically addresses the ease of rollout, configuration, and management of microwave and millimeter wave transport networks required to extend network coverage for wireless and 4G/5G end users. To that end it integrates applications on top of the open source OpenDaylight Controller that address key use cases in the domain – detection and configuration of new microwave devices, detection and operator-driven correction of discrepancies between actual and planned network configuration, visualization of configured topology, related and effective network capacity at each microwave link, notification of various network-related alarms and events, etc. The project enables these capabilities on a multi-vendor network of microwave devices from Huawei, NEC, SIAE, Ceragon Ericsson, Fujitsu, Nokia, and ZTE. They accomplish that via a piece of software-per-device called the Mediator that translates standard netconf into the device-specific proprietary language. Netconf manipulates the YANG configuration and state data that is auto-generated from the generic generic UML model of a microwave device (MW Model) that was developed based on ONF Common Information Model. The UML to YANG translation tool was a contribution of yet another OSSDN project, Eagle.
Project Snowmass drives interoperability and the benefits of SDN into the domain of transport networks, such as those that you would see in global tier 1/2/3 networks, mobile backhaul, carrier networks, data center interconnect, etc. These networks typically span multiple layers of transport (L0-optical channels, L1-optical transport networks, L2-carrier Ethernet, L2.5-MPLS-TP) technology. spread across multiple (metro access, core) geographic and administrative domains, and are composed of multi-vendor network domains and are currently managed via vendor-specific (EMS/NMS) software. With changing end user behavior, increasing demand on transport capacity and emerging new applications of transport networks, service providers need a way to quickly spin-up new services while keeping costs low and complexities manageable. SDN provides a path forward by abstracting away the heterogeneity of multi-vendor, multi-technology, multi-domain and presenting a uniform end-to-end network view to upper layer orchestration that OSS/BSS and/or new applications can easily integrate on top of. Project Snowmass builds an important component of that abstraction via its SDK that these orchestrators can use for programmatic access to the underlying multi-vendor, multi-domain, multi-technology transport networks. Today the SDK implements five common control plane services – topology service, connectivity service, path computation service, virtual network service, and notification service based on ONF Common Information Model. As shown in the figure, orchestrators can access these services via standardized APIs (TAPI) to control multi-domain greenfield/brownfield networks managed by different SDN controllers and/or traditional NMS.
If these projects spark your interest, find out more from Gordon Brebner (Project PIF), Martin Skorupski (Project Centennial), and Karthik Sethuraman (Project Snowmass) at the OpenDaylight Panel on ONF OSSDN, September 27, 2016.
- Bithika Khargharia, ONF OSSDN Director of Product and Community Management and Principal Engineer at Extreme Networks