OpenFlow with Broadcom OF-DPA Pipeline Switch Management Software for the B1
Line Rate SDN now at Standard Router Price Points
Software Defined Networking (SDN) promises to revolutionize the networking world. That promise is now being delivered by ZNYX with the LineRateOF switch management software. Powered by Broadcom’s OF-DPA technology, LineRateOF harnessing the power of their high-end, commercial switch silicon delivering the economics of scale that Broadcom switch silicon enjoys. Now SDN in general, and OpenFlow in specific, can be implemented on the same hardware, at the same line-rate, and the same cost structure as standard IP4/6 routing.
The “magic” that integrated the OpenFlow protocol with Broadcom silicon instructions is the OF-DPA Pipeline. The graphic below displays the table structure and process flow of the OF-DPA pipeline. The pipeline maps switch silicon data structures into OpenFlow programmable tables. Now OpenFlow instructions can be used to program Broadcom switch silicon. Learn More.
OpenFlow Throughput - Gigabits per Second
- Software OpenFlow Switch Implementations 7%
- Line-Rate OpenFlow for the B1 100%
- 480Gbps Switch
- 40G ports at line rate
- 10G ports at line rate
- Small 64Byte packets at line rate
OpenFlow with Broadcom OF-DPA Pipeline
- 120,000 L2 flows
- 16,000 L3 Unicast flows
- 4,000 L3 Unicast flows
- 4,000 Group Table Entries
- Up to 5,000 flow adds/sec
- Up to 4,000 flow mods/sec
OpenFlow 1.3 Compliance
- Action Sets
- Match Fields
- Multiple Tables
- Group Table
- Meter Table
- 480 Gbps switch fabric
- 24 10G ports
- 4 40G uplink ports
- 80 Gbps data path between server/switch environments
- Intel® Xeon® E5-2600v2 Processors with up to 24 cores
- Up to 512GB Memory
LineRateOF is available on the B1 Family of Platforms
The B1 Platform, which hosts LineRateOF management software, comes in two major platforms: as a switch only platform – the B1-n, or as a platform that combines the switch with a high-end NFV computer environment – the B1.
Both the B1 and the B1-n provide 400 Gig of line-rate OpenFlow to the ports on the front panel. The B1 provides and additional 80G of OpenFlow switching to the compute environment.
A standard Use Case for a B1 is deployment as a top-of-rack OpenFlow switch with network services applications running on the compute section. For example, a number of the compute section VMs could provide Load Balancing to some of the server in the rack while the other compute section VMs provide Intrusion Detection services for the rack. See the B1 product page for a full description.
The standard Use Case for a B1-n is as an OpenFlow switching at the top of the rack. This is the best OpenFlow price/performance option in the industry. A B1-n, with LineRateOF, out performs any of the software-based (OVS) OpenFlow switches by an order of magnitude. When the B1-n is compared to OpenFlow switches based on network, it delivers line-rate performance at less than half the cost.
The table below shows max flows, max table entries per second and max table modifications per second for the OpenFlow tables implemented by the Broadcom OF-DPA
|Table||Table #||Max Table Flows||Max Flow Entries / Second||Max Flow Modifications / Second|
|Ingress Port Flow Table||0|
|VLAN Flow Table||10||7,000|
|Termination MAC Flow Table||20|
|Unicast Routing Flow Table||30||16,000||5,000||4,000|
|Multicast Routing Flow Table||40||4,000||3,300||3,000|
|Bridging Flow Table||50||119,000||4,100||3,800|
|ACL Policy Flow Table||60||900||2,500||2,200|
|Group Table Entries||4,000|
OpenArchitect LineRateOF Software Stack
The core element of the OpenArchitect LineRateOF switch management software is the Broadcom developed OpenFlow abstract switch model – OpenFlow Data Plane Abstraction (OF-DPA). The Broadcom OF-DPA, based on the OpenFlow 1.3.4 specification, enables the B1’s switch silicon to be programmed as an OpenFlow Logical Switch using the OpenFlow protocol. ￼￼￼￼￼As shown in the graphic on the right the following components – OpenFlow Indigo Agent, Broadcom OF-DPA, and Broadcom SDK- are embedded within the OpenArchitect switch management of the B1.
The Broadcom OF-DPA Abstract Switch may be viewed as preconfigured and optimized to support single pass, full bandwidth packet processing performance that makes efficient use of the hardware and available table memory resources. This “fixed pipeline” trades off unrestricted generality in favor of latency, performance, and cost, while enabling a logically centralized control plane with programming flexibility.
The graphic below displays the table structure, and processes flow, of the OF-DPA pipeline. The pipeline maps switch silicon data structures into OpenFlow programmable tables. For example, the Bridging Flow Table maps into the MAC Learning Flow Table on the switch chip. This association provisions the capability for 120,000 L2 flow rules. Other key tables mapping into silicon data structures are the Unicast and Multicast Flow Tables. The Unicast Routing Flow Table enables 16,000 L3 Flow entries and the Multicast Routing Table enables 4,000 L3 flow rules. The ACL Policy Flow Table that enables 1,000 “wide” matching OpenFlow rules. The Group Table Entries enable 4,000 OpenFlow group entries.
Note that the OpenFlow Controller and Controller Applications have to be aware of the LineRateOF pipeline structure.
To make the learning OF-DPA curve easier, ZNYX provides working Application Controller examples that run on the B1. A REST interface with HTTP query example that directly interact with the OF-DPA tables are also available. The RYU Controller is used for this training and prototyping environment. The RYU controller is a Python based environment that is quick to setup and easy to understand. It is a perfect learning and OF-DPA prototyping tool.
Broadcom OF-DPA Network In-a-Box
The B1 can be configured as a stand-alone OF-DPA training and development environment. Installing OpenArchitect LineRateOF on the network environment creates a line rate OpenFlow switch. Installing Ryu in a VM on the Compute Environment with the direct Ethernet connection between the Network CPU and Computer CPUs used for the OpenFlow Controller link provides the controller-to-switch interface. Setup two more VMs – a traffic generator and client such as Apache Bench – on the Computer environment and you have a complete OpenFlow training and prototyping environment. ZNYX provides setup instructions for this environment.
Powerful Management Processor
The requests from an OpenFlow Controller to a switch can place demanding compute loads on the control processor. Tens of thousands of flow operations (adds, changes, or deletes) can appear over a very short time period. During such events processing these commands through the OF-DPA model to generate SDK API calls can generate heavy compute loads. To meet this challenge the B1 uses the powerful 4 or 8 core Intel® AtomTM C2000 Processor. This is a significant competitive advantage over the industry standard dual core processors used for switch management.Benchmarks of the performance of the ZNYX LineRateOF protocols stack show that it can execute up to 5,000 flow additions per second and up to 4,000 flow modifications per second.
Here is a list of links to key OF-DPA and OpenFlow 1.3 resource materials:
- OF-DPA Insider
A blog specializing in all things Data Plane Abstraction
- OF-DPA OpenFlow 1.3.1 Switch Pipeline Specification and Software
The datasheet for OF-DPA 2.0.
- OpenFlow Data Plane Abstraction (OF-DPA): Abstract Switch Specification v 2.0
Broadcom’s documentation of OF-DPA 2.0
- OF-DPA TTP OF-DPA
Table Type Patters in a JSON file. Useful to get explicit description of capabilities for a specific table
- OpenFlow Switch Specification – Version 1.3.4
The Open Network Foundation (ONF) specification for OpenFlow protocol, table flow entries, matches, actions, etc.