The URLLC debate (pt3): How will 5G slices be provisioned? (Will operators retain control?)
Network slicing, allowing for multiple network layers on top of a single shared infrastructure, is arguably the most significant capability of incoming 5G technologies. It brings brand new business opportunities to operators to serve enterprises and industries, each with conflicting requirements.
One of the panel sessions at URLLC 2018 in London earlier this month brought together specialists in certain of the key technologies that will enable 5G-based services – including slicing, virtualisation, edge computing, latency synchronisation – and notably the ultra-reliable low-latency (URLLC) variety of 5G communications that will underpin these use cases.
Xueli An, principal researcher at Huawei, was the resident slicing enthusiast at the London event. “It is not a technology, but an architecture,” she explained. In fact it might be better described as an architectural technique, or capability. Either way, it is a seminal part of the future mobile network.
“It allows operators to run many networks on the same infrastructure, with each customised for different industries. This way all their requirements can be served. It is the only way to do it with a single infrastructure.”
Its promise is great, and its theory is fine. The question, as with any game-changing innovation, is how it will work and disrupt things? Mansoor Hanif, chief technology officer at UK regulator Ofcom, and advisory board member at the UK government’s 5G support network UK5G, chaired the London event, and got to the practicalities of slicing early in the session.
“How will slices be provisioned – by vertical market, with slices available to the automotive sector or to manufacturing, say, or by use case in general, such as for low-latency communications? More crucially, who will provision these slices – operators, or governments, industries, and enterprises?”
That was Hanif’s big question on slicing for the expert panel in London. An at Huawei took it in her stride. Nothing is certain, she warned the industry’s old guard; everything is up for grabs. Operators may not hold charge of network slices, nor even command most revenues from this new technology.
“When the factory comes, the whole system will change. New rules and new businesses will emerge – we will have new partners, new customers, new models. To put it in a more provocative way: there is a big question mark, still, whether operators will even be the ones getting most revenue from 5G,” said An.
There is a powerful industrial set, she said, used to be in control, and a newer breed of service provider, running over the top (OTT) of broadband networks. In the end, the likes of Audi or Amazon will have scant regard for the old gods of mobile in pursuit of their own ends, she suggested.
“These OTT companies might buy a slice from an operator and integrate it into their own cloud offering, and serve customers that way. It is a really interesting phase. Operators need to grab the opportunity, and work out how to monetise it – because it’s really unclear who will make the money from 5G.”
Hanif restated the vision for clarity, and for full dystopian effect. “So you’re saying wholesale slice-provisioning could emerge, where operators provide [slices] to the likes of Amazon to serve [their own enterprise] customers?”
“There are all kinds of possibilities; not just this model,” responded An. “The industry has to evaluate all of these opportunities”.
What about the slices themselves? Will there be a handful of slices per network, or hundreds? “It is a typical question the verticals always ask,” said An. The latest 3GPP specification (Release 15) sets out three slice-types, for enhance mobile broadband (eMBB) and massive machine-type communications (mMTC), alongside URLLC.
“These three have very different performance requirements, and are fundamental. But there has been a lot of work on this, and we see that, based on current use cases, three is probably not enough.”
Discussions with the entertainment and manufacturing industries, for example, have raised questions about slices for multi-casting and security functionality, she revealed. “We are still on the way. But operators need practice time – the number will not be so high at the start.”
The operational costs go up with each slice, she noted, at least until operators have full network automation. Early on, the number of slices will depend in part upon the percentage of automation within the network. Operators will make the business case for each slice, weighing their customers’ demands and revenue potential.
Mansoor highlighted again the see-saw balance. “It’s a trade-off between agility and efficiency, against complexity and operations,” he said.
What about full automation? Will operators even achieve an autonomous system for network slicing? “Yes, definitely” said An. “But the deployment is out of step. That’s the reality. The current system does not allow for full automation. Because we have different technical domains.”
“We have to open up the ecosystem. Fully autonomous end-to-end network slicing has to come if we are to fulfil the vision. But it’s going to be a difficult path,” commented Mark Gilmour, director of portfolio strategy at Ciena, also on the panel.
How soon will operators get past this? “It is difficult to predict,” said An, reminding the room in London, again, URLLC capabilities hinge on upgrades to the end-to-end system, going beyond just the radio access network.
“It depends on the standardisation. 3GPP is only focused on the RAN part. The transport is important, as well. It requires 3GPP to talk with the other standards bodies to find common solutions. It comes down to the time it takes for the standards to come out.”
“There is the open source community, too,” said Mansoor, somewhat mischievously. Gilmour rejoined; remembering the response from the floor at an ETSI event in 2017, when UK telecoms provider BT stated in no uncertain terms it wanted total collaboration from the vendor community.
“It wanted fully end-to-end network slicing, and all the vendors to work together, from RAN through transport, through data centre, core, and into the application. And it was a controversial subject in there,” he said.
“And now? Is it still controversial?” Asked Mansoor.
“I would hope it isn’t. It’s a place we need to get to.”
As a footnote, Anthony Magee, principal engineer at ADVA Optical Networking – also on the panel at the London session, and featured elsewhere in this coverage of it – suggested the idea of full automation is not technically achievable in the short term.
“A percentage of resources in the network will be allocated to a fully autonomous system, but there will always be a niche application that needs absolute latency or reliability. And people will pay for it. And when that arises, you will want a hard configuration, as well as an autonomous one.”