This week starts the Open Networking Summit where Intel will demonstrate its support for the software-defined networking (SDN) movement with a demonstration of our Intel® Ethernet Switch FM6000 series working with an OpenFlow controller.
I wrote a post earlier this week that provides a good overview of the demo and Intel’s other activities at the show. Now, I want to give a little more insight into the technology that we’ve built into the FM6000 that makes it an ideal solution for these networks.
SDN technologies are revolutionary because they bypass L2/L3 IP protocols and instead use a software control panel and SDN data flow tables that direct incoming packets where to go. Thus, the switch is called upon to examine characteristics of incoming packets and switch them into an SDN-defined flow. As we wrote in a white paper that we recently posted:
Incoming frame header fields are looked up in the TCAM (ternary content addressable memory), which allows partial matches with the usage of wildcards. Matches in the TCAM point to an entry in the RAM, which controls various operations to be performed in the incoming frame header fields as well as extracting and propagating these fields to subsequent stages.
This logic can be built into the switch, but then what about any IP traffic that flows through that switch? One of the virtues of SDN is that it is a standard that can be used across switches from different manufacturers to “unlock” the network from proprietary protocols. But in the mean time, switch ICs need to support both traditional networking and SDN.
The FM6000 has built the parsing logic in its flexible frame forwarding unit so that it can be programmed to look for any networking protocol, either OpenFlow, standard routing or emerging data center switching protocols like TRILL, data center bridging and others.
The other advantage of this parsing flexibility is ability to deliver traditional Ethernet functionality that is not supported in OpenFlow, such as network address translation, load balancing and IP tunneling.
Of course, all of this must be done at wire speed and at very low latency, and that’s what you will see if you are able to stop by our demo at ONS. If you want a deeper look behind the technology, or take a look at our new white paper which gives an overview of our ONS features.
Software defined networking (SDN) has been called the next revolution in networking, because it promises to bring a better way to manage networks in a complex environment created by virtual servers and multi-vendor switches. It does this by separating the control and data forwarding planes and abstracting the control and management functions into a software-based controller that presents a centralized view of the network.
But that revolution has to start with the switch IC, which is the central piece of the network fabric and collects all of the data needed to manage these networks. At the upcoming Open Networking Summit, Intel® will be demonstrating how its FM6000 switch family is a key enabler for this next revolution in networking.
Traditional networking methods and protocols were acceptable for yesterday’s static networks, but managing complex and dynamic virtual networks has become extremely labor-intensive. Because of this, it has become far too time consuming and expensive to remain feasible and competitive in today’s virtual networks that are dominated with virtual device mobility, multi-tenancy and the need to quickly adapt to business needs.
Today, each switch in a network has a control plane that draws the network map and tracks the flow of packets through the network. SDN offers relief by creating a software-based controller that collects data from all of the control planes to provide a universal view of network conditions. OpenFlow is the leading SDN protocol, and it defines both the controller and the protocol that collects data from the switch IC.
Intel’s Barcelona 10GB Ethernet reference design utilizing Intel’sthe FM6364 switch silicon offers OEMs a way to support OpenFlow in high-speed, low latency data center networks, and we will demonstrate this capability at ONS (April 16-18 in Santa Clara).
In our booth, we’ll have a special demo that connects an OpenFlow controller to the Barcelona switch. Barcelona delivers full non-blocking performance on all 48 10GbE ports and four 40GbE ports with latency of 400ns for L3 switching. A 10GbE tester will provide the test environment and display how the switch and controller perform under maximum data rates.
The path ahead for the SDN / OpenFlow revolution runs right through the FM6000. In a future blog post, I’ll detail more about how the FM6000’s Alta architecture supports OpenFlow. But if you want to see it live, join us at ONS.
Arista Networks* has been a leader in the development of low latency 10GbE switches for the data center, utilizing Intel® Ethernet switch silicon. The company’s 7124SX switch is the lowest latency layer-3 10GbE switch in the industry, and they have now announced the 7124FX, which introduces a very clever new switch design to cater to the desires of their customers in the financial services industry (FSI) as well as other high performance computing applications.
Frame processing performance and latency are so critical in these applications, that traditional servers connected to the network through Ethernet controllers running communication software stacks don’t cut it any longer.
Arista has changed this by effectively embedding processing power in the network using an FPGA and CPU built into the 7124FX. In this way, packets can be inspected or modified as they fly through the switch at wire speed, minimizing the impact on latency as compared to a traditional network-adapter-server-adapter-network model. Financial trading companies can modify the FPGA code to implement their proprietary algorithms and a built-in atomic clock can be used to guarantee market fairness.
Intel Ethernet switching technology is at the heart of this new design. Not only do these 10GbE switches provide extremely low latency, they can maintain this latency while providing advanced L2-L4 forwarding features along with features such as data center bridging (DCB) and server virtualization support. In the 7124FX, Arista has creatively used an FPGA to add value, but in most designs FPGAs are used to implement network standards that are finalized after the switch design has been locked. Intel’s innovative FlexPipe™ frame processing pipeline is configured using microcode and can eliminate the need for external FPGAs in many advanced networking applications.