& cplSiteName &

Data Center Firms, Mobile Operators Pour Cold Water on Edge Computing

Mike Dano
5/9/2019
50%
50%

DENVER -- Big 5G Event -- The topic of edge computing has generated a significant amount of hype, and many in the space do agree it could play a key role in the ultimate development of 5G technology. But some top players in the mobile networking and data center industries are voicing serious concerns about edge computing in the near and even the medium term.

"Spend enough time in the telecom and technology industries and it becomes clear that the hype of many new technologies usually precedes the reality by 5-10 years. We believe that is the case with micro edge data centers," wrote Raul Martynek, the CEO of data center provider DataBank, in a lengthy post on LinkedIn.

Some speakers here at the Big 5G Event echoed that sentiment.

Jim Poole, VP of ecosystem business development at data center giant Equinix, said that mobile operators would need to completely revise their network designs away from voice services to get edge computing to work in a 5G world. "This whole thing needs to be changed, rearchitected," he said. "5G is an extraordinarily daunting change."

Poole likened the process to "turning around an aircraft carrier, in the mud."

And even Ibrahim Gedeon, CTO of Canadian mobile operator Telus, warned that there's a serious chance that the edge computing space could get bogged down in the telecom industry's standards process. "It took us forever just to come up with ORAN," he said of the wireless industry's work to separate vendors' various network components from each other, a goal many refer to as "open RAN." He said ORAN efforts would be great "if it was 1985."

Edge computing proponents argue that the mostly centralized nature of the Internet today won't support the snappy, real-time services that 5G providers hope to offer, like autonomous vehicles and streaming virtual reality. Such services require almost immediate connections between computing services and users, and an edge computing design would enable that instant connection by physically locating data centers geographically next to the users that need them. Such a design -- dispersed computing instead of consolidated in one location -- potentially could eliminate the tens or even hundreds of milliseconds it takes for a user's request to travel across a network to a computer that can answer it.

At least, that's the idea.

DataBank's Martynek argued that, at least so far, there's very little need for hundreds or thousands of mini computing locations spread out all over the country. Specifically, he noted that there are already several data centers physically located in most major metro markets in the US. For example, in Los Angeles, there are already two or three computing facilities that potential customers could use, he said.

But Martynek adds that a growing number of regional data centers can handle most edge computing needs today. For example, he said DataBank currently operates a data center in Minneapolis, which is roughly a thousand miles away from the nation's three main data center hubs: Ashburn, Va.; Dallas; and Santa Clara, Calif.

He said that one data center in Minneapolis essentially eliminates the need to build additional micro data centers in that city. "The incremental improvement from going from one data center location to 5 micro data center locations only improves your round trip latency by less than 1-2ms," he wrote.

As a result, he argued, "deploying in tens-hundreds-thousands of micro-data centers would only improve latency by 1ms or less, and in some cases introduce latency depending on where the peering occurs." Those comments essentially represent a dig at the likes of EdgeMicro and Vapor IO that are hoping to build out mini data centers in dozens of cities across the US.

Similarly, Equinix's Poole noted that edge computing is already available in a basic form today, considering that Equinix operates roughly 200 data centers around the world. And that raises the question of whether additional computing locations are really needed to support edge computing use cases like remote surgery or cloud gaming. "The classical mobile edge is a solution looking for a problem in some respects," Telus's Gedeon acknowledged.

However, most speakers agreed that, eventually, 5G would help spark more demand for edge computing services. "Does 5G need edge computing? I'd say the answer is yes. Does edge computing need 5G? The answer is no." Equinix's Poole said.

"I think edge computing is one of the two or three things that make 5G different," Gedeon said.

But Equinix's Poole argued that wireless networks need to essentially be redesigned in order to fully take advantage of the edge computing opportunity. Instead of routing all traffic through a handful of on-ramps, mobile operators will need to instead create ways for applications to immediately access local mobile users -- and to interoperate. For example, an autonomous driving system in Denver must be able to immediately route its traffic to 5G users in the city rather than through an operator's central routing location in Dallas -- and that system must also work regardless of whether the system uses AT&T or Verizon.

"This is real stuff. It has to happen. It can't not happen," said Poole, arguing that such complexity is akin to getting multiple technologies and companies to "dance on the head of a pin."

And Telus's Gedeon noted that, so far, there is no consensus about how 5G operators should design their networks to account for these kinds of edge computing use cases. "There is no fundamental way of rolling out the network," he said. "I fear we will all do our siloed things."

Mike Dano, Editorial Director, 5G & Mobile Strategies, Light Reading | @mikeddano

(7)  | 
Comment  | 
Print  | 
Threaded  |  Newest First  |  Oldest First        ADD A COMMENT
komatineni
50%
50%
komatineni,
User Rank: Lightning
5/9/2019 | 4:50:29 PM
Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
"Deploying in tens-hundreds-thousands of micro-data centres would only improve latency by 1ms or less, and in some cases introduce latency depending on where the peering occurs"

This summarizes the challenge with EDGE compute or MEC. Cisco started talking about FOG at least five years ago and delivered a hardware platform that can be deployed. Did it revolutionize the industry? no not at all. Reason: What and where is the business case. Just like with NFV, the MEC hype with AR/VR/V2X/Tele-X/Cache/micro EPC hit a wall and now the focus appears to be on latency. But the 'packets' need to be sent somewhere be it via peering or service termination. The latency is important and sub 10ms is critical for some very unique and niche services that are offered in specific locations without much mobility needs. V2X touted as the killer app but still going with 5G-G5 battle. So, in the end, it's back to basics and hope we'd go back to solving real challenges. 
Duh!
50%
50%
Duh!,
User Rank: Blogger
5/10/2019 | 11:11:01 AM
Re: Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
Propagation through fiber is a bit less than 100 µs per km, so 1 ms is a bit more than 10 km.  Does he claim that he has an existing data center within 10 km of every gNB (assuming that's where the micro-data center is collocated)?

But I agree that realistic use cases for <10 ms latency over 5G are few and far between.
brooks7
50%
50%
brooks7,
User Rank: Light Sabre
5/10/2019 | 1:20:11 PM
Re: Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
Duh!,

And that assumes that modulation delays and group delays are essentially 0.  This would mean that processing times are 0 as well.

seven

 
Duh!
50%
50%
Duh!,
User Rank: Blogger
5/10/2019 | 1:35:45 PM
Re: Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
That only counted first-order affectors. At >10 Gbit/s, all of those are in the noise.
brooks7
50%
50%
brooks7,
User Rank: Light Sabre
5/10/2019 | 1:52:39 PM
Re: Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
Duh!,

I think you will find compute time is actually longer than you think...especially if any scale of data is involved that requires retrieval from a disk.

seven

 
Duh!
50%
50%
Duh!,
User Rank: Blogger
5/10/2019 | 2:50:55 PM
Re: Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
Seven,

The 10 ms is just for the one-way network portion of latency. Compute comes out of another piece of the end-to-end budget.

Dan
Duh!
50%
50%
Duh!,
User Rank: Blogger
5/10/2019 | 3:40:26 PM
Re: Good one. Finally and hope the hype curve flattens soon and we can focus on real challenges and goals
I have to correct myself: Prop through fiber is 4.9 µs/km, not 98. Somehow my brain crossed Prop delay with attenuation. Max of 200 km to a data center is reasonable.

[imitates Emily Litella] Never mind.
Featured Video
Flash Poll
Upcoming Live Events
September 17-19, 2019, Dallas, Texas
October 1-2, 2019, New Orleans, Louisiana
October 10, 2019, New York, New York
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
December 3, 2019, New York, New York
December 3-5, 2019, Vienna, Austria
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events