The edge is much more than just a collection of connected devices
or sensors. As intelligence devices and sensors create massive amounts of data,
businesses must make technology decisions about getting the most value out of
that data. Is the analysis done on the device, edge, in the cloud, or at a data
center? Is the data retained for regulatory reasons or to find the root cause
of problems? Will new services like 5G play a role?
These issues are on the minds of many companies today. To
help identify key strategies and sort through the different approaches to edge computing
and the Internet of Things (IoT), we recently sat down with Krish Iyer, Strategy
Lead for Edge and ICV, and Calvin Smith, CTO, Emerging Technology
Solutions; both in the Office of the Global CTO at Dell Technologies.
We explored the role of emerging technologies, why edge and
IoT are so important today, factors to consider when deciding when to process
at the edge, compute considerations when using the edge, and what the future
holds. Here is a summary of our conversation.
Emerging
Technologies
RTInsights: What are today’s emerging technologies?
Iyer: We see a market inflection in several new
technologies. Edge is clearly a front runner along with IoT, AI [artificial intelligence], and ML [machine
learning]. Clearly, these technologies represent a market that is starting to realize
a great deal of traction. Most importantly, it’s the traction where many
customers are starting to see edge as an extension of their cloud, and they are
starting to look at edge as a way to distribute their workloads that cannot be handled
by some of their cloud infrastructure. That, in my opinion, is probably the
most important driver pushing many of the organizations to consider edge.
Smith: I think what’s interesting is this notion of
intersections. It’s not like edge emerged out of nowhere. Many of the early use
cases for IoT are now becoming more readily addressable because of the
combination of technologies. Krish mentioned IoT, Edge, ML, and AI. There’s
also 5G, Augmented Reality, and Virtual Reality. More broadly, as the costs
come down for compute and things like GPUs are used at the edge, the technology
capability goes up in terms of the amount of automation you can run.
It’s just amazing the proliferation of different
technologies that are being executed where the data is created, which is at the
edge. Ultimately at the end of the day, even if it [the created edge data] does
go back to a cloud or a data center, much of the actuation back to those
devices is going to be at the edge, too. It’s the net-new center of the
universe. It’s funny how things go in ebbs and flows, right? I mean, we went
from the advent of the PC and a more decentralized model to a monolithic data
center approach, then to a pseudo-monolithic cloud approach, and now we’re
going back to a distributed architecture. It’s really interesting to see how
things have evolved over time.
Iyer: Yes. I guess one way to look at Edge is that it
is a combination of heterogeneous systems. Edge is not monolithic, it’s not
homogenous, and it is a set of different functions. These functions are
typically used to collect data, to process data, and to store data. These functions
also need to transfer data to other functions or need to perform some action
based on that data to enable other functions like data processing, data analytics,
etc. These functions may also need to be performed in challenging environments,
like high-temperature environments, or rugged terrain. That’s why the term
heterogeneous is going to be very critical in edge.
Why All
the Attention on Edge & IoT?
RTInsights: You touched on this a little bit, but why
all the attention on edge and IoT, and why now?
Iyer: It’s interesting. As Calvin said, edge is not
new. Edge has always been there. And distributed systems have always been there.
But we are coming full circle again. The way the market is shifting is that
some of the functions or applications that typically ran on centralized data
centers or the core infrastructure are recognizing problems. Companies find
some of the applications have bandwidth or latency requirements that are not
achievable with a centralized approach. That necessitates moving these
applications closer to where data originates.
As Calvin said, moving the data processing closer to where data is created as laws of physics don’t permit running some of these applications at the core. It’s a speed-of-light problem you’re dealing with. It’s a natural phenomenon that’s causing the shift from the core to the edge.
See Also: Center for Edge Computing and 5G
Deciding Where to Process Data
RTInsights: That dovetails into the next question,
and that’s what factors should businesses consider when they try to decide
where to process the data? In particular, what considerations should be taken
into account when you’re trying to figure out: Do I process at the edge or not?
Iyer: Many of the factors are technical, but many also
could be business-related, as well as governmental- and regulatory-related.
From a technical perspective, again, the speed of light is a big factor. Even
if the cloud operators promise that they can process data at the core to
satisfy this requirement, the cost of doing that is going to be so high that
customers are going to say, “There’s no way I’m going to be able to pay those
costs. Customers are going to essentially look at this as a no-brainer. Move
the processing to the edge.
The second thing to consider is latency. For use cases like
autonomous vehicles (or AR [augmented reality] and VR [virtual reality]), it’s a
fact that a few milliseconds of lag is essentially the difference between safe
driving and an accident. For applications that require results in milliseconds,
or even faster, latency matters. To deliver the required results, the processing
has to be done at the edge.
The third thing I can think of is: how do you manage
bandwidth? The cost of sending these huge amounts of data that Calvin mentioned
to the cloud and back is going to be expensive and inefficient. It’s going to
drive up costs phenomenally, and that’s going to be a huge deterrent for most
customers.
Another factor that is important is security, especially
when it comes to edge. You can actually isolate some potential security
problems before attacks permeate into the core data center. Many organizations
can track and isolate some of the security attacks at the edge and close those
systems off before the problem comes to the core. You can actually do early
detection of any intrusion attacks, or any denial of services, and so on, right
at the edge before it reaches the core, shutting things off before an attack on
your central and your core infrastructure.
Then, there is the ability to scale. You’re looking at
environments where you can add additional sites or add additional environments
as your needs arise. If it’s a seasonal situation, or you just need to add more
functionality, edge provides high levels of scalability.
These are some of the high-level technical requirements, but
there also are things like regulatory requirements in the case of healthcare
applications or GDPR [General Data Protection Regulation compliance]. In most cases,
there are mandates that data needs to be collected at the location [where it is
generated], and not transmitted back to a central data center. Many organizations
must follow these mandates.
Smith: Krish
is spot-on. I would also add that it’s not like it’s a dichotomy that
it’s edge versus cloud or edge versus core. It’s a spectrum, a continuum. We
know that there are going to be workloads that run from the edge, others from the
core or cloud. It’s just a matter of placing the right workloads in the right
places; and executing against each at the right time. The notion a decade ago
was to collect and store everything, regardless of cost. Today, data is still key,
but it’s the analysis that adds the value – it’s been said that data is the new
bacon. Data is the new gold. Data is the new oil. That’s true, but not if it’s
static information that doesn’t add any value. What’s interesting is when you
start to do very basic filtering and machine learning at the edge. You don’t
have to send every instance of device data saying: “I’m alive, it’s 72 degrees,
it’s still 72 degrees,” back to the data center or cloud.
You don’t have to send those kinds of messages on a sub-millisecond
basis. If you do, it’s going to get very expensive very quickly when you look
at the sheer volume of devices in the world. You want to be able to parse that
data and make some sense of it at the edge, in situ. Some perishable, ephemeral
data only has value for a short period of time. What you really want to do is anomaly
detection to figure out what’s the important information we really have to send
back or keep. Back at the core or cloud, you can do your deeper analysis,
figuring out, how has this happened before? Is this anomaly happening to other
parts of the fleet of assets that we have in the field? That’s where the value
starts coming in. You need the whole stack and a singular view on the entirety
of your dataset. The important thing is, in addition to all the key parameters
Krish mentioned, there are also logical reasons that you need to consider.
There are parameters for your entire distributed architecture, and you’ve got
to figure out what makes sense to store, forward, analyze, and process, where,
when, and why. There’s a different logic for pretty much any architecture. It’s
all highly contingent on the use case and the infrastructure itself.
Considerations
for Moving Data to the Edge
RTInsights: Along those lines, what are the factors to consider for
shifting data and compute to the edge?
Iyer: The speed of light problem, costs, and security
are important factors. And we talked about bandwidth, high availability, and scalability
as other factors. The ability to reduce a great deal of data, to be able to do data
and metadata processing at the edge, and only send the most relevant data back
to the core is going to be another key factor. Much of this depends on the
vertical and the use case.
For example, telecom operators and content delivery networks
might have specific requirements for edge. They might need to leverage edge for
something deeper than many of the other verticals. These industries need to
figure out what kind of services to provide to users in a specific geographic
location. To do that, they might have to gather the context of the geographic edge
and be able to provide specific services for specific localities. The situation
might be different for, let’s say, operational technology use cases that
require doing predictive analytics at the edge for IoT devices and machinery. Again,
it all depends on and comes down to what specific vertical demands are.
For example, on the retail side, how do I make my customer user
experience really positive? How can I provide AR or VR experience that makes it
seamless, with no buffering involved? How do I make the overall user experience
positive and interactive, so the customer is able to make buying decisions
right there? Healthcare providers will have a completely different set of
requirements for applications like telehealth and other remote diagnosis applications.
There are also many regulatory requirements that come into play for such verticals.
Edge is so critical; it is something that must work.
Smith: We also need to expand people’s horizons on
what we use to define the edge. In an industrial context, an edge can be the
factory floor itself, and edge can be that car we’ve described as the moving
data center of the future. The car itself is essentially the edge. It could be
an offshore oil rig, the entire rig, or a section of it. It’s multiple things,
large and small, and wholly defined by the use case and what you are attempting
to do. Edge computing is the interesting part.
What’s also interesting is the form factor in terms of what
you actually use for edge compute – this is highly varied, too. Without being
too product-centric, Dell has gateways (which are very simplistic compared to
converged, or hyper-converged appliances) that do some of the protocol
normalization, some analysis, and can be used for some IoT platforms and
smaller form factor things. They have very finite and specific objectives and
map to a number of devices.
On the larger form factor side of things, we actually have solutions
called modular
micro data centers. We recently announced one, the Dell EMC Modular Data
Center Micro 415, a small, edge data center with built-in power and cooling and
remote management features. And we also offer one called the Dell EMC
Modular Data Center Micro 815 – essentially a full rack. These solutions
are flexible and scalable. As named, they are modular, and they scale up to
enable you to be able to build out your data center in a use-case defined,
composable way at the edge. We can literally airlift and drop in data centers
at your edge, regardless of the environment.
Think about that from a military context for people in the
field. Think about that for the top of a building where historically you’d want
to do the processing in a basement because there’s better cooling. Well, these
solutions have cooling built in. Part of the innovation is the chassis and the
enclosure and the way that it’s cooled and powered. We’re walking into these
new worlds where, to all the points Krish made about bandwidth, and cost, and
latency, there are also different constraints in those environments with regard
to vibration, and dust, and shock and hazardous conditions. We can literally
drop in ruggedized, enclosed micro data centers with storage, compute, and
networking that can solve problems in near real-time at the edge. It’s the
beginning of a very interesting change in the way people do business.
Edge and IoT Future
RTInsights: What do you see for the future with edge
and IoT?
Iyer: The enhanced need for all of the key points we
spoke about earlier will drive investment in the edge. Applications will drive
the future. It all depends on the type of apps and the developers that create the
apps. Applications are getting smarter every day. For infrastructure, or
environments to support these applications, they need to be smarter as well.
They need to grow at the same speed as the applications grow. Enhancements are
happening, and the industry is adapting to a disaggregated approach, by not
approaching it with monolithic infrastructure, and being able to right-size
themselves to support these applications. Yet, we must consider that the pace
of growth is not always there to support the growing demands of some of these
applications.
Besides being application-driven, the future of edge also will
draw large value from the cloud. Cloud is not going anywhere. I think cloud is
still going to be an integral control point for the edge. Cloud is still going
to serve as the key operating model, or the environment that essentially lends a
great deal of data processing, data handling, data management support.
Having said that, I think the future of edge is also going to be defined by how vendors come together. One thing that we’ve learned is there’s no single organization, or vendor, that has a monopoly over the edge. It’s a combination of multiple players that need to come together to provide services, hosting, operations, data, security, and so on, to a multitude of other vendors to form an ecosystem. This includes an ecosystem of proprietary vendors and an ecosystem of open-source vendors all coming together to provide end-to-end solutions. This cooperation is needed from application development to application support, to developer support, to security, to compliance, and more.
For the most part, edge infrastructure will be horizontal, and
vertical solutions will require systems integrators to bring it to the last
mile for verticals like healthcare, manufacturing, and military. That again is going
to require an ecosystem to come together to provide the needed functionality.
Smith: You mentioned the autonomous use case earlier.
I think that’s a really good one. The futurists are going to make provocative
statements and go, “Oh, edge is going to eat the world!” I don’t know if you
remember, but 10 or 15 years ago, everyone was saying, “Is cloud going to eat
the world?” Well, yes, to a certain extent, it did, but data centers didn’t go
away. At the same time, edge is not going to eat the world either. I mean, it’s
going to be big, and it is already growing, but cloud’s not going to go away.
If you look at an autonomous vehicle, it is kind of a car as
the moving data center of the future. I like this analogy because people
understand it. The car is essentially the edge itself, but then there are other
edges that it may be connecting to. There may be vehicle-to-vehicle (V2V) communications.
There may be vehicle-to-base-station or other infrastructure (V2X) communications
where you’re connecting to an LTE cellular service, or in the future, or
specific metros currently, a 5G service. Then, generally, there’s almost
definitely going to be cloud connectivity for things like fleet management and
looking at cross-connecting all these cars.
Again, it’s not cost-effective to send all the data from every
vehicle to the cloud, and it’s not fast enough for safety-related features if
you’re looking at things like smart airbag deployment, which is increasingly
getting smarter based on weight and different parameters of the passenger. Or,
object recognition for a multitude of cameras that have to enable an
autonomous, or semi-autonomous vehicle, or ADAS [advanced driver-assistance
systems]. All these place constraints and limitations. Basically, you have to
run some of this locally, but you also might want to dig deeper into the data
in the cloud or a core to get insights into anomalies, and that information is going
to influence other cars.
For instance, suppose I want to predict an airbag
malfunction or something that’s going to end up being a warranty recall. All
that important information needs to go up to the cloud, but again you don’t
have to send all of the static data – just the anomalous or error-recognition
data. Then there is tremendous value when you analyze this data across multiple
cars. Like you mentioned earlier Krish, an autonomous car is just a nice way of
thinking about tying it all together, but it’s not the only example. There are multiple
industries where you start to understand the harmony and rationale for placing
workloads in different areas across edge, core, and cloud. It’s an exciting
time. If we can do this same interview a year from now, and then ten years from
now, I would love to get back together and see if our edge proliferation prophecy
is accurate.
