The Intel® Development Forum in Beijing took place two weeks ago and the interest in my two SDN-related presentations was very high.

My poster chat drew about 40 or so people who stopped by in groups of 6-8 to hear the high-level overview to the Intel SDN story.  The attendance more than doubled for my conference session, where I went a bit deeper into Intel’s new data center and telecom network transformation initiative – giving a preview of the three product announcements made at the Open Networking Summit.

One challenge that is unique to China is scaling new web services for a potential market of 1.3 billion people – almost four times that of the U.S.  There were a lot of questions on this topic from top service providers, which I took to indicate that scaling is very important.

The other difference I noticed is that with less of a legacy network infrastructure than in the U.S., Chinese network managers are very open to trying new things to get the scalability and performance they need to deliver great service levels.

One key element of my presentation was a deep dive into why new networking platforms, like ONP introduced at ONS, are so necessary to advance the state of the art of SDN, and to providing ease of scaling in these high performance data centers.

As the server virtualization trend expanded into network virtualization, building high-performance, low-latency networks became much more complex for enterprises and data center operators.  New IP protocols like TRILL helped, but maintaining server/network coherency became very labor intensive.

To network managers dealing with this challenge, the SDN promise of separating the network control plan into a central network controller architecture was an immediate solution to a nagging problem.  And, first-generation networking products delivered on this promise by layering SDN onto existing switches. 

But the promise of SDN is much bigger than that; it’s nothing short of opening networks to a wave of innovation around new software functionality along with additional network cost-per-bit reductions. That’s the total story that ONP delivers on.

The potential of SDN for network innovation mirrors the transition from proprietary mini computer to the PC, which spawned countless innovations thanks to its combination of standard processors and operating systems and value-added applications.

In the network version of this story, enterprises evolve from vertically integrated networking platforms that are closed and slow to innovate, to a more open system with standardized switch silicon that has an open API to the control plane (or control planes for specialized applications).  These control planes then communicate through another API with apps running on a virtual server. 

This means that a network that had to be architected around special appliances to do packet inspection or provide security can now have those applications running on a high-performance server. The global controller will know what packets need to be processed by that application and will direct them to the application before forwarding them to their destination.

This architecture breaks down many of the barriers to entry in this market.  For new players, all they need are their software skills to develop their application. They can sell it into any network that supports the open API – regardless of the manufacturer.  On the other side of the coin, an existing software company can use standard hardware to easily develop its own complete solution, speeding time to market.

For every company that needs to scale quickly and keep network costs and complexity low – especially in fast growing economies like China – this is really good news.

The success of Arista Networks* is proof that there’s always room for an innovative start up – even in markets dominated by large players that execute well. But great innovation often requires market disruption to gain a foothold with customers.

In a recent interview with Network World, Arista President and CEO Jayshree Ullal talked about the market trends that helped Arista take off.  She said:

“Arista saw three disruptions in the market: a hardware disruption; a software disruption; and a customer buying disruption, which in my mind is the most important thing.”

Two of these trends are interesting to me because we’ve been participating in them.  First, the hardware disruption she mentions is the rise of merchant network switch silicon that has performance and features comparable to ASIC switches.

Our Intel^® Ethernet switch family is pioneering these merchant switch devices.  We not only provide throughput that is equal to or better than that of an ASIC, but our layer 3 latency is the industry’s lowest as well.

With a switch chip that provided competitive throughput and features to compete, Arista didn’t need to spend the large amount of resources on an ASIC development program, unlike some of its large competitors.

Instead, they were able to differentiate themselves with software – the second disruption on Jayshree’s list.  Arista developed its own operating system – the Extensible Operating System – leveraging Linux as the foundation.

Our Intel Ethernet switch FM6000 series silicon contributes to software innovation through its programmable FlexPipe™ frame processing technology.  FlexPipe’s configurable microcode allows switch manufacturers to update features or support new standards even on systems that are already in the field.

In order for our customers to evaluate the advanced FM6000 series features, we also provide our Seacliff Trail reference design, which has a Crystal Forest-based control plane processor on board.  Crystal Forest can be used as a standard control plane processor, or as a SDN controller host, or even to experiment with Intel’s Data Plane Development Kit (DPDK™).

It’s been great to have played a role in the market changes that have given Arista – and other companies – a chance to launch and to flourish.  Viva la Disruption!

What do hero pilot Sully Sullenberger, humorist Dave Barry and software-defined networking have in common? They all packed the house at the 2012 Gartner Data Center Summit event that I attended earlier this month. While Sullenberger talked about leadership and Barry kept it funny, Gartner analysts presented their research showing that a rethinking of data center infrastructure and operations can lead to dramatically reduced costs. Part of that rethinking includes adopting SDN

That got a lot of data center managers thinking and asking questions about what SDN is and what it can do.  At least that’s the response I saw as we staffed the Intel^® booth at the summit solution showcase.  We were there showing our Seacliff Trail (SCT) 10 Gbps/40 Gbps top-of-rack switch reference design along with our 10G Ethernet converged network adapters. You can read more about the SCT reference design here, which is based on the Intel Ethernet Switch FM6700 series.

The FM6700 series provides up to 72 10GbE ports or up to 18 40GbE ports and can forward frames at 960Mpps, while maintaining L3 latencies of around 400nS under all conditions. This product line is part of our FM6000 family, which continues our history of providing Ethernet switching silicon optimized for the data center. The 6700 series has been enhanced with advanced features for SDN such as large flow tables and support for VxLAN and NVGRE tunneling.

The data center managers I spoke with had heard Gartner’s message about reducing cost and improving network efficiency and had a lot of questions about how to turn the theory into action. This is another sign of the extreme excitement around SDN, and it was nice to see that many were becoming aware of Intel’s commitment to providing advanced SDN-enabled components.

Gartner itself is famous for its “hype cycle,” a graph that tracks the hype of a product over its lifecycle.  Exciting products emerge from a “technology trigger” and rise to the “peak of inflated expectations” before dropping in the “trough of disillusionment,” then emerging into the upward “slope of enlightenment.” In Gartner’s model, it's only after the products emerge from the trough that the market becomes real.

I’m not sure where SDN is along that curve, but after a few days at the summit it sure felt like the attendees were seeking enlightenment for how they could apply SDN in their data centers.

It used to be said that low latency networks weren’t needed for the data centers that ran big e-commerce or social media sites, as most people are willing to wait an extra few microseconds for the latest update on their friends and the network itself wasn’t a gating factor in performance.

Product recommendation technology, which powers the “you might also like” messages on e-commerce sites, however is changing that.  New “recommender” systems require increased computing performance to better factor more data into their recommendations.  And they need to do this all in the time it takes a webpage to load.

An article in the October 2012 issue of IEEE Spectrum by professors, and recommender system pioneers, Joseph A. Konstan and John Riedl chronicles the evolution of the technology and its dramatic impact on e-commerce sales.

The most popular recommender systems use either user-user or item-item algorithms – that is they compare your purchases, likes, clicks and page views with other people (user-user) or they compare the items you like with other items to see what buyers of those items also purchased (item-item). 

The two main problems with these approaches are that the algorithms are rigid and that tastes and preferences change, both of which lead to bad recommendations.

Dimensionality reduction is a new way to make both algorithms much more accurate.  This method builds a massive matrix of people and their preferences. Then it assigns attributes or dimensions to these items to reduce the number of elements in the matrix.

Let’s take food for example.  A person’s matrix might show that they rated filet mignon, braised short ribs, Portobello mushrooms and edamame with sea salt very highly.  At the same time, they give low ratings to both fried chicken wings and cold tofu rolls. The dimensionality reduction then seeks to determine that person’s taste preferences: 

“But how do you find those taste dimensions? Not by asking a chef. Instead, these systems use a mathematical technique called singular value decomposition to compute the dimensions. The technique involves factoring the original giant matrix into two “taste matrices”—one that includes all the users and the 100 taste dimensions and another that includes all the foods and the 100 taste dimensions—plus a third matrix that, when multiplied by either of the other two, re-creates the original matrix.”

So in our example, the recommender might conclude that you like beef, salty things and grilled dishes, but that you dislike chicken, fried foods and vegetables.

But the number of calculations grows dramatically as the matrices grow in size.  A matrix of 250 million customers and 10 million products takes 1 billion times as long to factor as a matrix of 250,000 customers and 10,000 products.  And the process needs to be repeated frequently as the accuracy of the recommendations decreases as new ratings are received.

This can spawn a lot of east-west data center traffic, which is needed to complete these large matrix calculations. Because users don’t spend much time on a given web page, the data center network latency is critical to providing recommendations in a timely manner (time is money).

Intel® Ethernet Switch Family FM6000 ICs are perfect for these types of data centers because of their pioneering low layer 3 cut-through switching latency of less than 400 ns.  So, the next time you get a great book recommendation, there might just be an Intel switch helping to power that suggestion.

Last week at IDF, we took the wraps off of two exciting new products that OEMs and ODMs can use to develop switch systems for the emerging market for software-defined networks (SDN) in the data center.

At the chip level, we launched the new Intel^® Ethernet Switch FM6700 series, which is a 10G/40G SDN-optimized switch family that provides up to 64 10GbE or up to 16 40GbE ports. 

The FM6700 series can support both SDN and legacy networks.  Thus it can be used in top-of-rack switch SDN applications in the data center, or in network appliances or video distribution switches (thanks to its built in load balancing features).

For all applications, the switch features a pioneering low-latency architecture, built on the programmable Intel^® FlexPipe™ frame-processing pipeline and single output queued shared memory architecture.  Both of these technologies combine to deliver highly deterministic packet forwarding with a maximum layer 3 latency of about 400nS.

The switch supports NAT and IP tunneling features for use in both IP and SDN applications.  For the SDN networks, the FlexPipe frame processor can be used to parse and process SDN packets.  The switch also supports 4,000 complete OpenFlow 12-tuple table entries that can be searched in a single pass for added performance.  There are also flexible tagging and tunneling options, including the ability to provide both an SDN and tunneling proxy for connected hosts.

At the platform level, we’ve introduced Seacliff Trail, a top-of-rack switch network reference platform for OEMs and ODMs that is based on the FM6764.  It offers 48 SFP+ 10GBE ports and four QSFP+ 40GBE ports, and can drive up to 7m of direct attach copper without the need for additional PHY chips on the board. 

It’s an all-Intel platform as well with a control plane based on the Crystal Forest AMC module that features an Intel^® Xeon^® processor, Intel communications chipset and Intel^® 82599 10GBE controller. Intel’s Wind River subsidiary provides the open and extensible software framework based on its Linux OS.  This provides both easy SDN integration and also direct API access to add third-party apps for rapid innovation.

Seacliff Trail is a major step forward in fulfilling Intel’s SDN vision of the next-generation of networking.  That vision combines standardized, high-volume hardware with an open and extensible software framework that allows OEMS/ODMs to add their own value-added functionality.

Last week we had good crowds coming to see these products at our IDF booth along with two sessions where we presented both the FM6700 series SDN features along with its features for server load balancing.. 

In my last blog post, I discussed virtualized network protocols NVGRE and VXLAN – two essential components in data centers that are transforming into virtualized environments. 

Another important component is balancing the traffic on each virtual server to optimize response time and overall resource loading.  Many data centers have installed expensive load balancing appliances in the network.  They are very surprised to find out that many of these same features are built into our Intel^® Ethernet Switch FM6000 Series products.  Here’s a little bit more about how it works.

Load balancing in the FM6000 Series architecture is done using advanced symmetric hashing mechanisms along with network address translation (NAT) to convert the IP address of the load balancer to the IP address of the virtual machine (VM) after determining the optimal virtual machine (or virtual service) to process the request.  After the transaction is processed by the VM, the load balancer modifies the source IP address to its own address so that the client sees it as a single, monolithic server. 

The FM6000 also provides fine-grain bandwidth allocation and fail-over mechanisms to each egress port using a flexible hash-based load distribution architecture. This avoids round-robin service distribution schemes, which may be less than optimal, and provides the ability to monitor the health of VMs and virtual services, so that failed ones can be quickly removed from the resource pool.  These switches also come with connection persistence intelligence to know when not to load balance, as in the case of FTP requests that must stay connected to the same virtual service.

Some other load balancing functionality built into the switches includes:

Network Security:  The frame filtering and forwarding unit (FFU) inside FM6000 Series can be used for network security, in addition to frame forwarding. It can be configured using bit masks to read any part of the L2/L3/L4 header. If there is a match, the switch can route, deny, modify, count, log, change VLAN or change priority of the packet to protect the network.  The switch can also use access control lists to prevent denial of service attacks and other security violations.

Performance: The FM6000 series switches are the lowest latency switches on the market, which means they can connect to the network, to servers and to storage arrays with real-time performance. In addition, it’s extremely low L3 latency means that the load balancing and NAT functions act as a “bump on the wire”, minimizing the impact on network performance compared to coupling a ToR switch with a discrete load balancer.

Fail Over: FM6000 series chips use a link mask table to determine how to distribute the load across multiple egress ports. They also contain several mechanisms to detect link failure such as loss-of-signal (LOS) or CRC errors. As the packet header is processed, the forwarding unit resolves to the address of a pointer, which points to an entry in the mask table. If a link or connected device fails, this pointer can be quickly changed by software so that the failing link is no longer part of the load distribution group. Since distribution is flow based, only flows to the failed device will be affected.

As you can see, the FM6000 Series switches have full-featured, low latency load balancing capabilities, another feature that makes them the ideal solution for top-of-rack switch systems.

Data center networking standards are constantly evolving – both for IP networks and for new software-defined networks (SDN).

This leaves most switch chip makers to either wait for the final standard to hard-code the protocols in their device, or to augment the switch with lower-performance programmable silicon and count on a software enhancement to boost speed after the chip has been designed into a system. 

Neither is ideal. That’s why we built innovative support for upgradeable microcode into the Intel^® FlexPipe™ packet-processing engine that is the foundation of Intel’s Alta switch chip architecture, the basis of the Intel^® Ethernet FM6000 family of low latency switches.  Microcode support gives FM6000 chips the ability to run new protocols with performance of a hard-coded solution and the flexibility of a software solution. 

Typically, when one hears the term microcode and programmability, it is assumed that the architecture is a run-to-completion or non-deterministic model.  FlexPipe operates in a deterministic manner, meaning that with any possible microcode implementation, the engine will maintain up to one billion packets per second of throughput performance and less than 400ns of L3 processing latency.

With the use of microcode, we can provide customers with a rich set of flexible features that can be adapted to changing market needs. Our customers can get to market early, before industry groups finalize standards, allowing future-proof system designs. It also allows customers to support standards that haven’t yet been introduced into the standards committees.

This means that the parsing and matching logic needed to forward SDN packets can be built into the switch today, as can some of the existing and evolving IP data center standards such as MPLS, IPinIP, NAT, EVB, VxLAN, NVGRE, FCoE or other proprietary switching headers for vertical markets. In addition, system administrators can test new SDN implementations while maintaining normal network operation with the use of two simultaneous microcode images.

FlexPipe is the key differentiator for the FM6000 family, giving them the performance and flexibility they need to deliver the top-of-rack switch performance necessary for high-port count virtualized networks.

One challenge for Ethernet in the data center is the increasing amount of “east-west” data traffic (i.e. server-to-server traffic), which demands much lower latency than traditional north-south traffic (ie server-to-user/Internet). 

Server virtualization and new web applications are spawning a dramatic increase in east-west traffic. For example, web content delivery can spawn hundreds of server-to-server workflows in order to provide a timely, customized experience for each unique client.

As we describe in a recent article in EE Times, iWARP technology combined with low-latency switching is the way to optimize Ethernet for these environments.

Historically, InfiniBand* has led a pack of proprietary, low-latency network protocols optimized for east-west traffic, delivering latency measured in nanoseconds from end-to-end vs. microseconds for traditional Ethernet networks.

The article shows results of tests conducted using the combination of the NetEffect™ Ethernet Server Cluster Adapters from Intel and the Intel^® Ethernet switching technology which showed latency performance that matched InfiniBand.

At the server level, the network adapter must speed the transfer of data from the server onto the network, a process that can add significant latency in normal Ethernet networks.  The NetEffect™ Ethernet Server Cluster Adapters from Intel support Internet wide-area RDMA protocol (or iWARP), a protocol that improves efficiency by eliminating the need to copy data from a receive buffer memory to server memory. Instead, iWARP has direct server memory access and doesn’t need to cache data in a buffer, which improves overall application performance.

Low latency is also important in switches, as even the most efficient data center network design involves two or three switch “hops” to have enough ports and bandwidth to connect all data center servers.

Intel Ethernet switch silicon provides low-latency Ethernet through its unique use of a true output queued shared memory architecture. By providing full bandwidth access to every output queue from every input port, no blocking occurs within the switch. In addition, since the packet is queued only once, cut through latencies of a few hundred nanoseconds can be achieved independent of packet size.

Data center bridging is being specified by the IEEE to enable converged data center fabrics. Intel switches support key DCB features and have an advanced TCAM-based classification engine that uses ACL rules to assign a traffic class to each frame. Based on traffic class, frames can be placed into one of several logical shared memory partitions in the switch and flow controlled separately. For example, iSCSI or iWARP traffic can be placed into one memory partition, while other data traffic can be placed into other memory partitions. This ensures that storage or HPC traffic will not be delayed or dropped if data traffic becomes congested in the switch.

Network performance has always been a critical element of data center computing, but the prospect of using the well-known Ethernet protocol in data center networks has a lot of benefits.  Delivering the necessary performance requires the right technology at the server and in the switch.  Intel’s combination of NetEffect adapters and Intel Ethernet switch silicon is the complete solution that delivers the promise of converged data center Ethernet.