Last week we analysed the memorandum of understanding announced between the two most prominent industry groups driving edge computing, ETSI’s MEC (Multi-access Edge Computing) and the Cisco-inspired OpenFog Alliance. This offered reasons to hope that this important area of technology for mobile operators would be on the road to greater unity. Like ETSI’s other major initiative, NFV, fragmentation has been threatening to derail MEC, but even now that it is working more closely with OpenFog, there are many other groups and approaches vying for attention.
This is threatening confusion, frustration and delays among operators, many of whom are unconvinced of the business case for MEC, and daunted by the costs it will entail. That debate, in turn, is threatening to undermine the case of many of the low latency, edge-oriented applications which were supposed to be a key justification for 5G.
All this uncertainty could deter mobile operators, and that would risk opening a door for alternative providers of edge-based services, such as Amazon AWS with its Greengrass platform. Faced with this situation, a few telcos may even seize the initiative themselves, as AT&T did in the area of NFV management and orchestration (MANO), when it open-sourced its ECOMP technology.
In the MEC area, the operator making waves is Deutsche Telekom, which has launched its own edge computing initiative called Low Latency Prototyping. This will have its kick-off later this month and has emerged from DT’s incubator program, hub:raum, which has a program focused on industrial robotics, virtual/augmented reality, drone steering, retail automation and self-driving vehicles.
“If we talk about low latency computing we need to manage performance end to end,” said Thomas Gerszberg, SVP of business operations for group innovation, and also DT’s representative on one of the many edge computing projects, OEC (Open Edge Computing). “I believe it will come in Deutsche Telekom’s network very soon.”
This may be primarily an internal project so far, but Low Latency Prototyping is addressing some of the issues which MNOs and telcos think ETSI and others should be handling more rapidly. And the structure of the DT effort, via incubated start-ups, would lend itself to being expanded to third parties and even open source, like Facebook’s Telecom Infra Project (TIP). Telcos, including DT, Orange, BT and AT&T, are setting up labs and start-up programs around TIP, driving an ecosystem which has the potential to be a full alternative to the industry of Ericsson and Nokia.
The shift of the mobile and converged network to being a distributed cloud/apps platform should also help with this disruption of the telecoms industry, introducing new concepts and new suppliers, and a hefty dose of the open source tradition. But the major downside is that too many companies may be battling for a piece of the market, and that leads to the dreaded fragmentation, and to a focus on warring specifications rather than a workable business model.
Gerzberg told the audience at last week’s MEC Congress in Berlin: “From the operator perspective the first question is how I can sell this great asset that is edge computing. So far you are not helping a lot because there are so many acronyms… It is natural that we ended up with many concepts and many competing solutions, because, if you get money for innovation, you need to promise that you will improve your position for this innovation.”
Even the various organizations involved in defining edge computing are concerned. “There is more work that needs to be done on interoperability testing between these different systems,” Steve Vandris, Intel’s director of IoT and 5G, and a board director of the OpenFog Consortium. Told the congress. “We need to create open interfaces to ensure there is interoperability. Interoperability testbeds are the next goal.”
Of course, ETSI and OpenFog are working on shared APIs and other cooperations now, but that will not be enough to create a unified approach, even without the Open Edge Computing initiative (driven by Dell) lurking in the shadows, not to mention edge-oriented efforts within TIP. Each of these efforts is based around a different concept, so cooperation will not just be a matter of agreeing some core interfaces, it will about embracing one another’s viewpoint.
That will be an essential process if the telecoms and IT industries really are to converge around the ‘edge cloud’ and provide a fully connected, fully distributed environment on which any kind of application and vertical market can run. This not easy though:
ETSI shows its mobile telecoms roots with an approach which relocates IT resources, such as processing and storage, close to the user. ETSI MEC early movers like Nokia have envisaged the resources being collocated with small cells or other cellular elements, though they have pulled back a bit in recent times to support a more flexible view of MEC resource location.
OpenFog has a broader view in which computing nodes are distributed everywhere in the telco network, from the core to the actual devices. At the very edge, Cisco has outlined ‘mist computing’ for an even more dispersed vision. In fact, some MEC participants, such as virtualized EPC supplier Quortus, have also pushed edge computing resources beyond the cell and into devices, gateways, enterprise controllers and vehicles.
Open Edge Computing (OEC) arose from Dell’s work on edge-based gateways, and so has IT roots like OpenFog, and a heavy focus on the IoT, which will, of course, be a major driver of the shift to the edge. ETSI and the OEC have been cooperating over the past year.
Then there is the CORD initiative (Central Office Re-architected as a Datacenter) part of the Open Networking Foundation and with a separate stream for mobile networks, M-CORD. As its name suggests, this open source initiative is looking to convert operators’ central offices into data centers equipped with white box hardware and open source software but it is also taking its own approach to the edge. This is attracting considerable interest from many telcos, as is Facebook TIP, which is increasingly focused on edge networking technologies, working on this with BT and others.
These two groups make the open source approach prominent, and that presents certain challenges for ETSI, in keeping its more conventional processes in the forefront of MNOs’ minds.
There is definitely a shift of thinking towards a fully distributed, rather than just edge-based, cloud. Gunnar Bergman, a technical manager at ARM, told the MEC Congress that
mobile network-centric approaches, like MEC, were just relocating resources from the center to the edge, not distributing them where they were most needed. “You cannot just have computing close to the base station. You need to be able to distribute to where it makes most sense and that is what OpenFog is about – putting compute and storage where it is most useful.”
Rival Intel has a similar view, and Vandris said: “The difference with fog is that the computing resources are distributed throughout the network. Each fog node can offload to neighbors in all directions. It is a truly distributed hierarchical cloud.” A node can live in a device or in the central data center, or any points in between.
Gabriel Brown of analyst firm HeavyReading believes these initiatives can work together if they regard ETSI MEC as an underlying infrastructure and are able to write to its MEC APIs. He said: “From the MEC perspective, OpenFog is an application, while from OpenFog’s perspective MEC is about infrastructure. If OpenFog can write to [ETSI] MEC’s APIs, it can carrying on developing its own system architecture but make use of the ETSI environment.”
Frank Michaud, a member of the OpenFog Consortium and a technical leader at Cicso, took a similar line at the Congress, saying: “My view is that we have a precise fog architecture and on to that we are mapping the APIs of MEC and will do that with other consortiums so that hopefully we get one story. We need a unified story on fog and edge.”
But despite recent alliance announcements, many at the MEC World Congress were sceptical. A Huawei delegate complained: “When we talk about interoperability a good starting point would be where it is needed. What are the reference points and interfaces? Who is in charge of defining something in that particular interface?”
Another important step, for adoption of these ideas within 5G, will be acceptance by the 3GPP standards body. Reznik is confident that 3GPP will adopt ETSI’s MEC APIs, telling us: ” I expect they will expose services in a way that is complementary to what MEC has done. When they go to APIs I hope they will follow our guidelines as different services are exposed.”
One of the key reasons why 3GPP and its members are interested in edge computing is that ultra-low latency services are supposed to be a significant justification for 5G. The promise is that 5G networks will support far lower latency, lower power transmissions than 4G, supporting a wide range of IoT and M2M applications which were previously impossibly on cellular links. To minimize latency further, it is important that a great deal of processing of IoT signalling and data is done very close to the user or ‘thing’, rather than everything having to be sent to the cloud or core.
But in reality, how important and practicable will ultra-low latency services be in early 5G, and will these really be a major driver of 5G investment? Most first-stage 5G deployment plans are heavily focused on conventional mobile operator services – boosting data rates for consumer video and data applications, and even for fixed wireless. And if MNOs are uncertain about the case for ultra-low latency, that will also weaken the case for edge computing investment. Of course, MEC supports many other use cases, with video caching being one of high relevance to the core enhanced mobile broadband business case for 5G, but that can be done with other, possibly cheaper, techniques in existing infrastructure.
Its ‘killer argument’ is to enable brand new services and revenue streams with demanding latency requirements. Without that element, it will be far harder for most operators to justify the cost and disruption of entirely re-architecting their network around distributed data centers, integrated with their connectivity.
That cost could be very high, Kye Prigg, Vodafone UK’s head of mobile networks, has identified edge computing and backhaul as the highest costs of 5G. And a new report from analyst firm iGR predicts that telcos and enterprises in the US and Europe alone will spend $272bn on edge computing capabilities between now and 2026. That will be a bigger sum than for 5G networks (less surprising, since it includes enterprises and fixed-line operators, not just MNOs). I
ain Gillott, president of iGR, sees every commercial building having a MEC node of some kind, but one of the key questions will the same one which has dogged small cell deployment. Will these be 5G-connected, and if so, will they be operator deployed and funded? Or will the enterprise pay for the MEC capabilities, which would reduce the MNO’s cost, but would deny it the opportunity to have those MEC nodes as an integrated part of its cloud, supporting revenue-generating services?
In many cases, enterprises or local fixed or private network operators will be able to harness MEC and shared spectrum, along with small cells, to create self-contained local clouds with cellular connectivity – linking to the wide area MNO network just for long distance roaming. This would remove a hefty element of cost from the MNO, but would also risk the operator being squeezed into a long-range bitpipe role.
But even if the MNO controls a lot of the edge, it still has to work out how to monetize that beyond just connectivity fees, which in the IoT are likely to be very small. And only a few ultra-low latency services have yet been identified as having near-term revenue potential at a higher level than connectivity – industrial control and real time video at sports events are much discussed, while others, like robotic surgery, are a distant prospect.
As operators grapple with these business dilemmas, and hope for a unified technical approach, there is a sense that ETSI’s mobile-centric vision is being superseded, and that it made a wise move to come close to OpenFog last week. But that burst of unity will not be enough to create a fully harmonized edge platform for operators, and far more work must be done to avoid fragmentation and create a more compelling business case for a 5G world based around low latency and the edge.