Executive Summary:

There were many significant announcements at this year’s Open Networking Summit (ONS), held June 15-18, 2015 in Santa Clara, CA.  Unlike in past year’s conferences, there were no vendor sales pitches or infommercials.  That was certainly refreshing!  However, ONS continues to focus on “pure SDN” (as per the ONF definition) vs. various other SDN reference architectures, such as network virtualization/overlay model.  That’s largely because the ONS is closely aligned with the Open Networking Foundation (ONF) and ON.Lab (which developed the open source Open Network Operating System). Surprisingly, there were several companies (Google, Microsoft, AT&T) that said they were using OpenFlow (but not how) in their proprietary versions of SDN.

THE MAJOR THEME OF THE CONFERENCE: Most SDN and NFV SOFTWARE IS GOING TO BE OPEN SOURCE – from organizations like OpenDaylight, ONOS lab, OPNFV, and ONF.

The open source movement will drastically disrupt many companies that had planned on their own vendor specific implementations of SDN Controllers, Open Flow, and NFV virtual appliances.  The key missing piece of vendor neutral SDN and NFV software is MANAGEMENT & ORCHESTRATION (which includes scheduling and service insertion/chaining).  Until that’s provided by one or more open source communities, proprietary implementations will inhibit widescale deployment of SDN (and even more so) NFV.

I believe the Atrium project (see description below) is a milestone for ONF in that it expands their scope to include open source software for SDN.  Atrium will likely complement Open Daylight and ONOS open source projects. It’s still very early in the SDN Open Source space while OPNFV (for NFV) is in its infancy.  Some efforts will succeed; others will fail. IMHO, commercial, supported Open Source SDN/NFV code will be critically important to the open networking industry.

Key Take-Aways:

• The ONF’s Atrium open source software was demo’d at the ONF SDN showcase on the exhibit floor.  ONF Executive Director Dan Pitt described it’s significance and ongoing trend setting work within ONF during an interview with this author.
•  ONF also demonstrated OpenFlow hardware interoperability using SDN data plane switches from five different vendors (Accton, Quanta, NoviFlow, Netronome, Corsa, Pic8, and Centec).  They achieved this by defining a mediation sublayer called “Flow Objectives,” which the data plane switch companies I/O driver software/firmware interfaces with. Hence, it masks the hardware differences/uniqueness of each data plane switch vendor’s product.

• ONOS, the community developing the open source SDN network operating system and control plane, announced that software’s first wide-scale production deployment on the Internet2 research network. It demo’d an open source SDN router running on white box hardware.  The Internet2 deployment is the first big deployment of ONOS on a live network.  ONOS sponsors fund the open source software development wihich is done in Menlo Park, CA by programmers that work there.

• Dell and Huawei expressed interest in the commercial open source business for SDN with Huawei favoring ONOS for teleco/carrier SDN and OpenDaylight for enterprise SDN as well as smart cities and IoT.
• John Donovan, AT&T’s senior EVP for technology and operations said the company only uses open source now for ~5% of its software, but would like to increase that to 50% in the next few years.  He said AT&T’s “secret sauce” (proprietary) software should be used sparingly, “like tabasco sauce.”   AT&T is an active participant in OpenDaylight and Open Platform for NFV (OPNFV). AT&T is also an active participant in TM Forum’s ZOOM program, which is actively building relationships with both OpenDaylight and OPNFV.  AT&T is contributing a design tool to OpenDaylight that the company has been using it its quest to “disaggregate” its network as part of its Domain 2.0 initiative. That means disentangling all the various subsystems and stripping down to core components, Donovan explained. “We have to rethink how they’re constructed,” he said.  Note that AT&T is not, and has never been, a member of the ONF.
• AT&Ts goal is to convert 75% of the network equipment hardware boxes in its network to software running on generic/commodity compute servers by 2020, according to Mr. Donovan.  From a base of zero, 5% of AT&T’s network equipment will be implemented in software by the end of 2015, Donovan said.
•  AT&T is on an “aggressive schedule” to make the optical transport layer more flexible, according to Andre Fuetsch, AT&T’s senior vice president of architecture and design.  “We want to start pushing NFV down the stack. We want to push it to Layer 1,” Fuetsch said during a keynote panel at the Open Networking Summit (ONS) on Tuesday.
• CORD for GPON: AT&T, along with chip vendors PCM-Sierra and Sckipio Technologies, are using the ONOS SDN software on a proof-of-concept project called Central Office Re-architected as Data Center (CORD).  The basic premise of CORD is to transorm AT&T’s Central Offices (COs) so that they look like cloud resident data centers with compute servers doing all the heavy lifting.  AT&T plans to dis-aggregate the functions of each network element- in this case a GPON Optical Line Terminal (OLT) – so that as many software functions as possible to run on generic compute servers. The white box hardware for CORD-GPON is a “stackable OLT” with GPON blades that are connected via Ethernet to a top-of-rack switch.  Here’s a paper on the CORD fabric.an open source leaf-spine CLOS fabric 
• CORD for g.fast: AT&T also demo’d a CORD proof of concept project for G.fast.  An OpenFlow-enabled G.fast distribution point unit (DPU) was connected to a G.fast CPE bridge from Sckipio. [G.fast is an ITU-T DSL standard for short local loops (<500 meters) with performance targets between 150 Mbit/s and 1 Gbit/s,]
• For the first time anywhere, Google disclosed its internal data center network architecture and Jupiter jumbo data center switch. Scale out, performance and availability are Google’s key challenge in network design.  The company designed their own routing protocol and didn’t use any open source software. Google’s current data center network design has a maximum capacity of 1.13 petabits per second using merchant silicon with their own routing protocols and software.  For comparison, the current data center throughput is more than 100 times as much as the first data-center network Google developed 10 years ago. The current network is a hierarchical design with three tiers of switches, but they all use the same “off the shelf” chips. Google’s control software treats all the switches as if they were a single switch.
• The Jupiter jumbo switch leverages the latest generation of merchant silicon, has 80 Tbit/s bandwidth, and uses some form of SDN (Google proprietary) with OpenFlow used in an undisclosed manner.  

Google’s Jupiter “cluster switches” provide 40 terabits per second of bandwidth—about as much as 40 million home internet connections. SOURCE: Google.

• We strongly suggest the reader view/hear Amin Vahdat, PhD keynote on Google’s evolving data center network architecture..  Amin is one of the best NO NONSENSE presenters on networking that this author has ever observed.   His keynote speech is here.  We also recommend you read Amin’srelated blog:  A look inside Google’s Data Center Networks
• Microsoft Azure public cloud uses its own unique version of SDN, which was described by CTO Mark Russinovich during his keynote speech.  Fifty-seven percent of the Fortune 500 use Azure (we find that remarkable as Amazon’s AWS is by far the #1 public cloud service provider  for Infrastructure as a Service -IaaS).  Microsoft’s cloud storage and compute usage doubles every six months, and Azure adds 90,000 new subscribers a month, and this places unprecedented demands on its network, Russinovich said. The number of host computers quickly grew from 100,000 to millions.  Microsoft  Azure needs a virtualized, partitioned and scale-out design, delivered through software, in order to keep up with the explosive growth of users and data.   Microsoft uses virtual networks (Vnets) built from overlays and Network Functions Virtualization (NFV) services running as software on commodity servers. [Note that virtual networks and overlays are NOT permitted in “pure SDN” and are a completely orthogonal SDN reference architecture].  

–Vnets are partitioned through Azure controllers established as a set of interconnected services, and each service is partitioned to scale and run protocols on multiple instances for high availability.  

-SDN Controllers are established in regions where there could be 100,000 to 500,000 hosts. Within those regions are smaller clustered controllers which act as stateless caches for up to 1,000 hosts.

-Microsoft builds their SDN controllers using an internally developed the Azure Service Fabric,  which has what Microsoft refers to as a “microservices-based architecture” that allows customers to update individual application components without having to update the entire application.  Microsoft’s Azure Service Fabric SDK is available for download here.

-Microsoft Azure’s SDN doesn’t use any open-source software. Russinovich said that’s because open-source communities don’t provide the functionality Azure requires.

-Russinovich said Azure’s SDN gets a hardware assist from a SmartNIC FPGA.  SmartNIC covers those functions that need a hardware boost, or that Microsoft would just prefer to offload from the CPU.  User data traffic flows through the SmartNIC – for functions such as encryption, quality-of-service processing, and storage acceleration. “The sky’s the limit, really, with what we can do with an FPGA given its flexible programming,” Russinovich said.   He said that compute servers are better at running virtual machines to serve Azure customers, rather than to take on mundane network processing tasks that could be better implemented in silicon.  Mark’s keynote can be watched here.

—>NOTE that this is the exact opposite of AT&T’s position on disaggregating network element functions and moving as much software as possible to generic compute servers, leaving only physical layer transport in network equipment.  Microsoft evidently believes that hardware assist from smart NICs is still a valuable proposition for SDN.

• Service provider adoption of SDN/NFV: From Talk to Action (?): Andre Fuetsch of AT&T, Kang-Won Lee of SK Telecom, and Yukio Ito of NTT communications  shared their works, experiences, challenges and roadmap in the keynote panel, moderated by Guru Parulkar, Chair ONS, at the ONS-2015.  The common aspects among all the three telcos were: (a) Importance of Open Source (b) Presence of both technical and non-technical challenges in adapting SDN/NFV (c) Experience in real-world deployments of NfV use-cases, and (d) Importance of role of SDN in transport networks, especially optical transport. 

—>This author noted that the “attack surface” exponentially increases with SDN and NFV so asked a question on mitigation of cyber security attacks.  As that issue has been “swept under the rug” for a long time, it seemed to stun the moderator with no clear solutions from the three carrier panelists.   Checkout the video clip yourself to see if the answers were “on the mark” or not.

• Alibaba disclosed the China eCommerce giant was using BOTH the pure SDN/OpenFlow and the Network Virtualization/Overlay model in the same network.  Panel moderator Guido Appenzeller, PhD of VMWare said that was the first time he’d heard of both SDN reference architectures being used within the same network.  How do they inter-connect, if at all?

• During a Thursday, June 18th Opening Keynote Panel titled SDN in Enterprises,  NSA’s Brian Larish said, “centralized control via OpenFlow is key” and “we are all in on OpenFlow.”

• Cavium’s XPliant® Switch was the winner of SDN Idol 2015, the open networking industry’s top award for the year’s hottest SDN solutions. The four finalists for SDN Idol 2015 were: Cavium, XPliant Switch; Huawei, 2015 SDN IDOL – DN-based IP & Optical Synergy; NEC, SDN Cyber Attack Auto Protection; and Pluribus Networks, Integrated Network Analytics.

Selected Quotes:

• “Embrace Open Source SDN or become irrelevant going forward.”  Guru Parulker, PhD and Chair of the ONS (each year since its inception)

• “Open Networking is here now and encompasses disaggregation of networking technologies including hardware and software similar to Servers 20 years ago. We expect the momentum of ON/SDN/NFV to pick up in 2015 for both Carriers and Enterprises and we are excited that Dell was a leader in creating and embracing this disruption.”  Arpit Joshipura, VP of Dell Enterprise Networking & NFV and a long time colleague who this author truly respects.

• “SDN and NFV are speeding up innovation, as seen in projects like CORD,” said Tom Anschutz, Distinguished Member of Technical Staff at AT&T in a press release. “These technologies create systems that do not need new standards to function and enable new behaviors in software, which decreases development time. Faster development time leads to rapid innovation, something the industry needs to continue satisfying data-hungry customers.”

• “Data Center Anchored Communication Services (DACS) is one of the key motivations for using SDN….We are all on a leash when we use the public network.  Sometimes we like it because we get “move around” with IP, sometimes we would have wanted more bandwidth. But there is a leash tied to an anchor for each communication service – statefull inline control.”  Sharon Barkai, founder of Contextream which is now part of HP. [Note that ConteXtream has historically gone after service providers to deploy its software into data centers and centralize management of networks.]
• “The (networking) industry is porting anchors from network appliances, and Application Specific Integrated Circuits (ASICs), to software on standard (x86) compute servers. Getting these DACS to factory mode in a row of clusters involves flows, granularity, federation, overlays, mapping, Overlay Descriptor Lists, etc. The challenge is to map traffic to computed anchors dynamically per flow. Virtualization frees any given hardware from ties to specific service functions and specific subscribers.  This is where the capex (utilization) and opex (factory) savings come from and how feature velocity enabled.  SDN really determines what gets processed where,  and therefore, done right, holds the key.” Sharon Barkai of HP (see previous quote).

• “Open-source may, in the long-run, be more secure, but it’s a tough one to buy. I guess one would have to look historically at open-source software projects and ask if the mere fact that a project was open-source contributed to its robustness, or was it that the robustness was introduced by someone who took the open-source effort and really provided that additional software that made things secure.” Vishal Sharma, PhD, Principal Metanoia, Inc & IEEE ComSocSCV officer.

• “Software Defined Networking is now migrating from a discussion to actual implementation in the real world….The XPliant protocol independent programmable (Ethernet) switch silicon is providing a platform for innovation to further this reality.”  Eric Hayes, VP/GM, Switch Platform Group at Cavium (winner of SDN Idol award as noted above).

• “This year’s ONS is rife with SDN and OpenFlow solutions such as the Atrium Project that demonstrate how these technologies provide a whole new tool-set for solving network problems. We now face the challenge of making the benefits this technology known to a wider audience, and affecting a truly fundamental change in how networking is done! It’s a fascinating time to be in the networking business!”  Marc LeClerc, VP Strategy and Marketing at NoviFlow Inc.  [Marc described the Atrium OpenFlow (ONF control plane to 5 different data plane switches) interop demo to this author and Ken Pyle – see above illustration].

Acknowledgement:

The author would like to thank Elise Vue, Publicist for EngagePR who was in charge of the media for this conference. Unlike most PR agencies that ignore or neglect accredited media, Elise went out of her way to ensure I was getting everything I possibly could from ONS 2015 each and every day (Tues, Wed, and Thurs).  It was a truly refreshing and positive change from most conferences I’ve attended recently.  Elise also followed up after the conference to provide feedback on the unscripted video interview I did with ONF’s Dan Pitt:  “I agree with Dan, it was seamless! Great work, keep it up.”  THANKS for your excellent support, Elise!

References:

Alan’s interview with Dan Pitt, Executive Director of ONF:

How ONF is Accelerating the Open Software Defined Networking (SDN) Movement

ONS Video Archives:

http://www.archives.opennetsummit.org/archives-ons/archives-2015/