OpenFlow on Advantech ToR Switch

Isn’t innovation in today’s network  accelerating at such an exciting and incredible pace?  Much of  the work  so far has beeen focused on data center traffic engineering with emerging use cases evolving very quickly now in service provider networks. Everyone is talking about Software Defined Networking (SDN) along with Network Functions Virtualization (NFV) and these are what are fuelling all the innovation.

SDN fundamentally allows network administrators to manage network services through abstraction of lower level functionality. This is done by decoupling the system that makes decisions about where traffic is sent (the control plane) from the underlying systems that forward traffic to the selected destination (the data plane). For the control plane to communicate with the data plane  some  mechanism or protocol is required and so far, OpenFlow has been the defacto mechanism employed by most developers and manufacturers.

OpenFlow v1.0 has actually been used in commercial networks for a while, and for various reasons including lack of open platforms at initial release time,  relies on a single table model based on Ternary Content Addressable Memory (TCAM) or ASICs. Solutions built this way are expensive from a hardware design perspective, especially when we scale up the network topologies and the number of flow entries which need to be  supported.  Things will ease off somewhat when OpenFlow v1.3.1 is released as it enables the use of multiple tables in packet processing pipelines which will help OpenFlow v1.0-based solutions to scale more cost-effectively.

In a recent blog, Paul Stevens described what was going on over at the Advantech booth during Interop 2014 in Las Vegas, where we took two major steps forward with our SDN strategy. Paul described the SDN and NFV demonstration on Advantech Carrier Grade Servers using a flow processor based NIC and software from Netronome.

In this post I’ll describe our second demo,which was in fact a showcase of Broadcom’s OpenFlow Data Plane Abstraction (OF-DPA) software running on our ESP-9212 Top-of-Rack switch which we announced at the SDN World Congress in 4Q 2013.


Advantech ESP-9212 Top-of-Rack Switch (Front View)

Broadcom’s OF-DPA software is currently based on OpenFlow v1.3.1 and addresses the challenges of cost-effective scalability. It provides a hardware abstraction layer (HAL) for Broadcom StrataXGS switches, and also provides an API library for OpenFlow 1.3.1 compliant agents. It is intended for both real network deployments and enablement of  research networks. The Advantech ESP-9212 switch is available with Broadcom OF-DPA as an easily deployable development package to enable network innovation.

OF-DPA OpenFlow objects are programming points for Broadcom network switch devices. These include flow tables with action sets, group table entries, logical and physical ports, meter tables, and queues. OF-DPA manages hardware resources and states on behalf of OpenFlow. In addition, OFDPA supports OpenFlow specific state such as aggregate counters and flow entry expiration. Broadcom leverages the hardware functions inside StrataXGS switches to provide the besy possible performance.

The chart below shows the OF-DPA Abstract Switch Pipeline for enabling L2 bridging + L3 routing. OF-DPA flow tables accommodate specific types of flow entries with associated semantic rules, including constraints such as which match fields are available, which instructions and actions are supported, how priorities can be assigned to flow entries, which flow tables entries can go to next, etc. The flow tables fully conform to the OpenFlow 1.3.1 specification. OF-DPA provides API calls to support interrogating tables for capabilities including supported match fields, actions, instructions, etc. They also include status properties such as current resource usage. More detailed information can be found at

Broadcom OF-DPA Abstract Switch Pipeline


What makes the ESP-9212 different from “white-boxes” which are generally available in in the market, is Advantech’s  use of an Intel Xeon processor as the Local Management Processor with support for up to 32GB DDR3 memory —  this can provide much more computing performance for virtualized networking functions.


Advantech ESP-9212 Top-of-Rack Switch (Inside view from rear)

In addition, multiple  ESP-9212’s can be linked together  logically as one single switch, running the applications on the top. This provides many benefits for managing the whole network environment.

Typical applications which can be added to the switch  include Virtual Tenant Networking, Network Virtualization,  Traffic Engineering  to  increase application performance, Service Chaining and Network Monitoring.

One of the key messages we were trying to get across at Interop to those who stopped by to see the demo was about Advantech’s flexible engagement model. In addition to it’s standard switch design Advantech is engaging heavily with network equipment manufacturers on many customized or semi-customized switch designs where we can offer  the experience we’ve gained from separating the data and control planes in hardware. This means mixing and matching different levels of processing performance with scalable interconnects, switching silicon and  I/O ports.  We’ve been doing this  pretty much since our group was formed, however over the past couple of years this has taken many more shapes and forms: from 1U/2U systems to standard and customized eATCA designs with terabit switching capacity.

We call this the Advantech NCG Customized COTS framework and it is growing strong with much greater flexibility to help equipment manufacturers accelerate their network platform evolution. Some examples  of Customized COTS in network switching beyond just branding and logistics include:

1U/2U system design tailored to meet specific needs:

  • Enterprise and Carrier Grade
  • Redundant / non-redundant AC and DC power supplies
  • Single or redundant fans, fixed or hot swappable
  • Lights out or advanced remote management

Choice from a range of ready to use Intel Architecture platform IP:

  • Multicore Atom and Single Socket E3 Xeons for cost optimized OpenFlow Implementations
  • Single and Dual Socket Intel Xeon E5 Series processors for additional service and application capabilities

Scalable interconnects:

  •  PCIe for switch management
  • GbE for OpenFlow Control
  • 10G/40G for user traffic

Switch silicon building blocks that meet customer specific performance / price points

Customizable front I/Os for 1G/10G/40G/100G

Check us out at, add a comment below or drop us a line at we’ll make sure your switched to the experts.

Terry Yang – Product Manager Advantech USA




Leave a Reply

Your email address will not be published. Required fields are marked *

5 × 5 =