Carriers, fixed and mobile, are embracing network virtualization with almost alarming speed. Fully ratified formal standards will take another couple of years, but many operators are banking on the de facto specifications, from various industry groups, becoming sufficiently concrete offer the confidence factor of unified standards. Most important is ETSI’s NFV (Network Functions Virtualization) initiative, and several carriers are already running extensive trials and evaluating the range of use cases, with a view to commercial implementation from next year.
However, while industry-driven specs move more rapidly, and more directly in response to carrier needs, than formal standards processes, they are also susceptible to the influence of major players with their own agendas. When it was first proposed in 2012, NFV was seen by many as a way for large network equipment suppliers to seize control of the wider moves towards virtualization and software defined networking (SDN). Virtualization up-ends the traditional profit models of the proprietary equipment makers, by running network functions as virtual machines on standardized hardware, and it provides a wide-open door for IT-oriented vendors like Oracle and Dell to enter the previously rarefied world of carrier platforms.
Operators, having been stung in the past by the need of their suppliers to control platforms, seem determined to keep the initiative this time. Most of the major virtualization trials, whether using NFV or inhouse approaches, have been widely multivendor, as seen at Telefonica and China Mobile. And the related, but wider, shift to SDN is also allowing leading proponents to extend their pool of partners, as AT&T has done by inviting companies like Metaswitch and tail-f into its inner circle (though Cisco quickly countered by acquiring tail-f). This shows that the greatest defence carriers have against a new generation of platforms monopolized by a few giants is to insist that everything remains open source.
In many ways, open source processes of various kinds are taking over the role of traditional standards bodies. They can be just as susceptible to the political and commercial interests of a few power players, but they are more fluid, more transparent and, of course, do not carry the hefty costs of participating in a full standardization activity. The shift to core platforms which are based on open source technologies will be as important to the carriers’ new network and cost models as the move to virtualization, and the two will often go hand in hand. Software-based networks will be the most significant architectural shift of the coming decade, and a strong opportunity to do things in new ways.
This was highlighted by the leaked news that a group of technology companies held a meeting in Silicon Valley this week to kick off the creation of an open source NFV implementation. As reported by LightReading, this group ‘ named Open Platform for NFV or OPN – also pointed to another important trend in modern network platforms ‘ the convergence of fixed line, cable and mobile architectures around common approaches in which the different providers can share R&D and accelerate progress towards fully seamless IP. For instance, the new NFV initiative was reportedly hosted by CableLabs, the organization which oversees standards for that sector, but includes executives from telcos and cellcos too, including AT&T and NTT Docomo.
The aim is to create an open source reference platform, containing both hardware and software, to validate the performance and interoperability of NFV components and to attract inputs from the open source community. Its founding document, seen by LightReading, points to a pragmatic approach, saying: “The overall design should be modular, to allow for the existence of proprietary components. Initially, the focus will be on two critical components: NFV Infrastructure (NFVI) and Virtualized Infrastructure Manager (VIM).” So proprietary, as well as open source, components will be able to gain OPN compliance certification, but the reference hardware will be based on off-the-shelf elements.
And the OPN is not setting itself up in opposition to ETSI’s initiative or to conventional standards bodies, saying these are important enablers for NFV too. The first release of specifications from the ETSI NFV SIG is scheduled for the fourth quarter and this work will be complemented, not sidelined, by OPN. The SIG will “remain the main hub of influence for NFV-related activities’ focusing on ‘use cases, requirements, and architectural aspects that define what is needed’, says the document. Meanwhile, the OPN’s main role will be to take those requirements and specs, and develop the reference implementations and the testing and integration systems.
Another important role for the OPN will have to be to unify open source efforts. The biggest disadvantage of open source approaches, as seen time and again in enterprise and wireless technologies, is the potential for fragmentation and for variable quality of implementations. That has often led to a single vendor seizing an initially open platform in the interests of bringing consistency and economies of scale (Google is still struggling to deal with the trade-offs between unity and freedom in Android). If an operator-driven group, rather than a single giant, can be that unifying force, the NFV movement will benefit significantly. And indeed, the OPN says it will work with all the open source communities relevant to NFV, including the Linux Foundation (which will host the OPN work), OpenStack, Open vSwitch and OpenDaylight. The aim will be to avoid duplication of effort and to provide an integrated platform.
One of the open source initiatives with which the OPN will need to cooperate is likely to be Metaswitch’s Project Calico, launched this week. Although the brainchild of a single vendor, the project is intended to become part of OpenStack, which is increasingly seen as the most promising option for NFV virtual networking. And Metaswitch has a strong track record of showing how open sources processes can boost the impact of a small vendor (it is now part of AT&T’s SDN supplier roster).
About a year ago, it open sourced its core IMS code and put it in the cloud, in an initiative called Project Clearwater ‘ a somewhat similar effort to that of CloudNFV, set up by consultant and campaigner Tom Nolle to demonstrate virtualized functions and increase confidence in cloud-based approaches. Last year, Nolle called on ETSI to “forget the notion of designing NFV to be capable of running on everything from bare metal to the cloud, and adopt the presumption of cloud implementation. That would simplify the process of specifying how functions could be deployed and operationalized, and take advantage of the enormous body of work cloud computing has produced.” ‘(Nolle’s latest project is an open source management and orchestration model named ExperiaSphere. This aims to orchestrate physical and virtual resources, provide bridges to existing BSS/OSS systems, and expedite service delivery.)
Meanwhile, Metaswitch’s new activity aims to move the virtual networking component of NFV from Layer 2 to Layer 3, which would simplify implementation and drive internet scale. Current implementations of OpenStack run at Layer 2, which entails the complexity of using Ethernet vLANs or other overlays and tunnels, in order to support legacy enterprise protocols. That involves a lot of “wrappering and un-wrappering’, says Metaswitch CTO Martin Taylor. Layer 3, by contrast, is simpler and more efficient, and that will be important as NFV deployments grow in scale and spread beyond a single data center. Metaswitch is offering Project Calico under a free software licence via Apache, and it will take over the networking function of the default OpenStack system when deployed.
However, Metaswitch is also experimenting with more lightweight approaches to NFV than full OpenStack’style virtual machines. Taylor says there is rising interest in lightweight containers which do not virtualize the full machine image, complete with operating system, but support many instances of an app, but riding on the host OS to save resource. Calico would be able to unify functions running in full VMs with those in lightweight containers, if the latter concept takes off.
These examples show how many open source groups are being formed around NFV, and only a few will gain the big-name backing to achieve critical mass. The success of open initiatives relies on broad support, so how wide is the OPN’s appeal likely to be, considering operators and vendors are already influencing NFV via the ETSI group?
Some vendors will see this open source effort as anathema, but others will welcome it as a way to lower their own barriers to entry to new segments. Ironically Intel, the largest chipmaker in the world, falls into this category in telecoms, but now has the golden opportunity to shift network functions to its x86 server platform ‘ a scenario in which it has nothing to lose by supporting the most open standards possible, if they look likely to accelerate that transition. Intel, then, is a founder member of the new group, as is Hewlett-Packard.
The x86 server community will benefit from openness, but has its own challenges, primarily to convince carriers that cloud servers can really deliver the same performance for network functions as specialized and optimized equipment. These players will welcome the OPN’s commitment to a reference platform based on standard, high volume hardware, but the group will reflect the complexity of the network issues better if it also attracts more traditionally carrier-focused chip and equipment suppliers. While many of the systems which operators will feel confident to virtualize early ‘ such as IMS and security gateways ‘ may run quite naturally on x86 servers and even in the cloud, the deeper into the core and access networks they go, the more they are likely to need a more variegated model. Nicholas Ilyadis, CTO of Broadcom’s infrastructure and networking group, hit the nail on the head when he said recently that operators would need a hybrid approach to NFV, combining commodity servers with dedicated appliances. Of course he has an axe to grind, but he is right that while x86 or ARM servers will bring low cost and reusability, some applications will require the throughput of specialized kit, and the commodity servers will perform best if they offload intensive tasks, such as packet header processing or encryption, to dedicated accelerators. “You can take a server and run it as a network switch and get 40Gbps throughput,” he said in an interview, but a commodity switch can deliver 32 ports, each of 40Gbps, adding up to 1.2Tbps at the same wattage as the server. “If you have the switch do what it does well, and have the hardware do what it does well, and bring the two together, you have a more flexible implementation,’ he pointed out.
The key to openness is not to run all functions on the same hardware, but to ensure that there are standard APIs (application programming interfaces) for specialized accelerators.
As for operators, they have their own profound cultural shifts to address in order to flourish in the open source world. While many carriers are always happy to wait for a standardized platform to appear before deploying new technology, there is an important group which drives innovation and takes on the risk of creating and rolling out cutting edge systems. These operators often have highly secretive R&D operations and, while they donate some of their developments to standards bodies such as the 3GPP, they also regard differentiation in their network performance as a competitive edge.
So carriers will need to learn to be open too, rather than just demanding that their suppliers are. A hopeful sign was seen when Diego Lopez, a senior technology expert at Telefonica’s R&D division, told the LightReading Big Telecom Event that several major operators were planning to link up their NFV reference labs and share their R&D findings, in order to accelerate progress ‘ and ensure carriers stay at the helm of NFV evolution.
Telefonica has been one of the most aggressive telcos in testing virtualization, from core to RAN to CPE, and hosted some impressive demonstrations at this year’s Mobile World Congress. The Spanish giant has created an NFV reference lab based around commodity switches, hypervisors, OpenStack virtual infrastructure management and orchestrators. It is now “federating reference labs” with other providers to “validate common use cases, experiment with collaboration mechanisms, explore business models, and build a global reference implementation’, he told the conference, though he did not name any of the partners. However, he did say that the labs would collaborate via a virtual network based on Alcatel-Lucent’s CloudBand platform, another indication of how that vendor is playing a strong hand in virtualization. It is a key partner in China Mobile’s huge Cloud-RAN trials, for instance.
ALU may well prove to be an example of a traditional vendor which, under intense pressure in its established markets, will successfully reinvent itself for a more open, cloud-based world. One thing seems certain ‘ no supplier will get away with trying to harness NFV in a proprietary way any longer, and successful vendors will have to learn to work with open source groups and operator-led initiatives, not just the familiar standards organizations. This is not just a power struggle, but a recognition by carriers that their processes and supply chain are far too slow and cumbersome to allow them to respond effectively to over-the-top giants which harness the internet model. As Lopez summed up:”We want to shorten the roadmap, and break the cycle of long term development.’