Highlights of Light Reading’s White Box Strategies for Communications Service Providers (CSPs)

Abstract:

Network operators (like AT&T, Orange, SK Telecom, etc) and mega-web companies (like Linked In) are moving away from proprietary purpose-built hardware to “white box” solutions that are a mix of purpose-built software, or open source software running on generic compute server and/or Ethernet switch/IP router hardware.  A “white box” is to be distinguished from a “bare metal switch” which does not come with any software.

It’s been said that white box networking offers greater flexibility and the potential for faster service delivery, but that’s yet to be proven.  The transition from dedicated network equipment to generic white boxes is one part of the transformation process being undertaken by communications service providers and large enterprises as they seek greater efficiencies and new business opportunities.

The path to a white box world is complicated. In the past, the white box approach was only considered viable for enterprise networks, data centers and, in rare instances, the very edge of service provider networks. The latest iterations reach across vast swaths of wide area networks (WANs), right up to the carrier core, and the closer service providers get to an end-to-end, white box “white wash,” the higher the stakes become.

Inline image

On November 17, 2015, Light Reading sponsored an excellent conference on “White Box Strategies for Communications Service Providers (CSPs).”  We report the highlights in this post.


Key Points Made by CSPs:

Ken Duell, Assistant VP of New Technology Product Development & Engineering, AT&T said there were many potential benefits of white boxes, including cost per bit, flexibility, speed to market and operations costs savings.  
“When we open up our network (to white box vendors) that leads to innovation.  Some of our customers come up with things we never even thought of,” he said.
 
However, there are huge challenges which include integration of white box hardware and open source/ purpose built software as well as network operations. Currently, AT&T is doing the integration themselves, rather than outsource it to a systems integration 3rd party.  “Right now, because of the state of the ecosystem, we’re doing it ourselves,” Duell said. AT&T  would like to involve a whole ecosystem, but for now it looks to best-of-breed components.  The telco giant (which is really SBC which acquired the old AT&T) is retraining its employees to handle the shift to white boxes/ virtualized network functions.
 
[LinkedIn is also doing the systems integration themselves.  They say they can save on OPEX by disabling network functions that are not needed at configuration time.]

AT&T’s first step in white boxes was what they call “universal CPE,” which the company describes as follows:
“Universal Customer Premises Equipment (uCPE) is a virtual appliance. Utilizing network function virtualization (NFV) technology, uCPE network appliances can be run as software on a virtual machine. Instead of installing a physical router, a customer could turn up a software-based virtual router in near real-time.   Not only does the Universal CPE afford increased speed but, for the first time, the ability to run multiple functions simultaneously. The unique open design enables multiple virtual network functions (VNFs) to run concurrently. For example, if a customer wants a router and a WAN accelerator, they can activate both of those functions on the same physical box.”

 

Next up is the CORD Project for GPON network equipment disaggregation.  CORD uses ONOS– an Open Source SDN operating system for service providers.  Announced this past June, a CORD field trial will take place in Q1-2016. Mr. Duell described the project as “an open, virtualized service platform that provides cloud economics and agility. Think of it as “FTTH as a Service” consisting of hardware blueprints (schematics?) and open source software modules, Mr. Duell said.

From AT&T’s website: “CORD enables service providers to build an underlying common infrastructure with white boxes using ONOS (carrier-grade open source SDN Control Plane), OpenStack (virtual infrastructure management), and XOS (an open source service orchestration/management platform built on OpenStack) with a diversity of organizations building the services and solutions that ride above. In effect, this common infrastructure replaces the fragmented, non-commodity one in today’s Central Offices where each site hosts more than 300 unique deployed appliances, each requiring a physical install and specialized management.”

Guru Parulker, PhD and ‎Executive Director of the Open Networking Research Center (ONRC) at Stanford University, said CORD was just one of several solutions enabled by ONOS, which is gaining a lot of momentum on a path to real deployments.  He didn’t disclose any other solutions.  Both ILECs and CLECs could take advantage of ONOS, Guru added.

Christos Kolias, PhD and a senior research scientist at Orange Networking Labs in S.F. is a frequent speaker at Silicon Valley networking conferences related to virtualization technologies.  This new world has a 3 layer reference model vs the 1980’s OSI 7 layer model, according to Christos.  The 3 layers are:
– Applications: encompassing L3-L7
– Control layer -network software: SDN, NFV, network virtualization, network OS, etc 
– Data layer- infrastructure: physical & virtual (L1-L2)
Note: We would put L3 = Internet Protocol (IP) in the Control layer for route/path computation AND also in the Data layer for packet forwarding.  That’s independent of a centralized SDN controller is used for all route/path computations in a given network domain.

Kolias sees integration as the key challenge for communications service providers (CSPs) that want to adopt the same kinds of hyperscale technologies that are fueling their biggest competitors, namely the web giants like Amazon, Microsoft, and Google.  He also said that white boxes pose a huge challenge for traditional network equipment vendors, such as Cisco, Juniper, Alcatel-Lucent, Huawei, and Ericsson.  “The user (network operator) now has a choice.  Reusability will be important; software will be the differentiator.” 

Christos identified the opportunity to “create a marketplace for services and applications” including emerging Internet of Things (IoT) and marketing apps such as trading advertisements for free bandwidth to support video streams, as well as the ability to build customized products to the specifications of the service provider, as Google and others do today.   All of that is enabled by separating hardware and software, and moving to more commodity-type, commercial-grade hardware built on merchant silicon that can rapidly scale and support innovation at the software layer. Not that getting to that stage will be easy, Kolias said.

Noting that the CSPs need “disaggregation, not disintegration,” he said there are significant integration challenges to achieving the same level of consistent performance using white boxes that exist today with purpose-built telecom gear.

– Breaking down the network gear

– Systems Integration/packaging options: ODM (h/w or s/w), 3rd party, end-user, or a combination thereof

– Accelerates product development to match evolving requirements

– DevOps ‒ continuous testing & integration

– The rise of the ISVs (Independent Software Vendors)

– White box switches and software switches

– Suitable for Data Centers, Enterprise, Cloud, potentially CSPs

As noted above, there are different approaches that network operators can take to address the systems integration (hardware, open source software, purpose built software, etc) challenge, the Orange executive said: CSPs can rely on hardware or software vendors for help, bring in third-party integrators, do their own integration, or combine those three approaches.

APIs are more critical than ever, Christos said.  They are important for plug-n-play, especially for open platforms (Google, FB, Microsoft, eg, WebRTC).  APIs provide: the “glue” in an open system, the integration points with existing systems (eg, OSS), the support points for the apps.  APIs are essential for building developer communities.

– Standard & open APIs can enable and ensure interoperability and thereby lessen dependency risk 

– They can enable plethora of innovative (eg, ad-hoc/customized) services and lead to new business models for the telcos 

– Provide monetization opportunities (eg, consumers, enterprise, VNOs, etc)


Ashish Singh, SK Telecom’s GM/VP of products said the South Korean telco is transitioning from an ARPU business model to an Average Revenue Per Service model.  SK Telecom is planning to use white box servers as part of a strategy to build “mini-modular data centers” for services at the edge to serve mobile computing needs, rather than sending data to a centralized data center.

The goal is to move intelligence closer to the edge of the network, to enable improved reliability, data protection and end-to-end encryption, Singh said.  He added that IoT services via NFV (Network Function Virtualization) must provide: low latency, high reliability, processing that’s close to the user.  He cited telemedicine as an example IoT application.  At the mobile edge, SK Telecom is using Apache Spark for real time streaming video.  Again, the objective is to push decision making to the network edge.


OPEX Savings or Problems?

I continue to believe that OPEX will be a key issue for white box networking because the white boxes won’t be as reliable as typical carrier class network equipmen.  Furthermore, the hardware/software are from different vendors leading to a systems integration/fault isolation/finger pointing problem which will increase OPEX.

Key OPEX issues include: hardware/software integration, fault isolation and repair (fail-over and problem resolution), restoration/re-routing around failures or congested network nodes, integration with OSS/BSS, upgrades/updates to software which cause problems with other functions, etc.


Conclusions:

1.  The AT&T and SK Telecom panelists said they are using the same infrastructure (network equipment) for both wireline and mobile subscribers via a new wireless access rack (not defined and no details provided).  

2.  While CAPEX for white boxes will be lower than purpose built networking gear, we think OPEX may be considerably more expensive (see section above).  This point can only be proven or disproved several years after white box networks have been deployed in large volumes.

3..  A huge transition is underway for telcos – from a network operator offering a fixed set of services to a “platform service provider,” which permits new companies to build applications and business models that run on top of the network.  Such new age CSP’s Comms will open their network and resources and build their own services on the platform, while also letting partners build services as well.

Comment: Haven’t we heard that line before? Like every other day for the last 15 years?


 ODMs that make White Boxes (Compute Servers and Bare Metal Switches):
Quanta, Accton, Delta Networks Inc, Edge Core, Foxconn, Super Micro, ABMX, Servers Direct, AMAX, etc

Software Vendors & Systems Integrators that use White Box Servers and/or Switches:
Cumulus Networks, Big Switch Networks, Pica8, Centec, NoviFlow

References: