IOT INFRASTRUCTURE

  • 11/11/2015
    7:00 AM
  • Rating: 
    0 votes
    +
    Vote up!
    -
    Vote down!

IoT: The Enterprise Impact

The Internet of Things will place new pressures on many fronts for enterprise networks, including DNS and management requirements.

A lot of people believe the Internet of Things creates challenges due to the additional number of devices being placed into the network.  While it's true that more devices place stress on the core infrastructure, IoT is more complex than just having more computers or mobile devices on the network.

Preparing enterprise networking infrastructure for IoT isn’t just a matter of pulling more cables, putting in more WiFi access points and WAN links, and it's not like supporting BYOD. It requires addressing increased DNS demands, new protocols, and different management requirements.

How IoT is different

Let's start by looking at how IoT is different from regular Internet connectivity, which for the purposes of this article is defined as a connection between people using devices, which are connected to apps.  IoT is broader, connecting objects, apps, business processes and people through the data shared between these items.  These range from devices on people (wearables, health care sensors), sensors in the real-world infrastructure (parking meters, road sensors), and vehicles (airplane engines, fleet of delivery trucks, trains).

IoT devices’ behavior differs from people (or apps like browsers) accessing the Internet.  When you use a web browser to access a web page, the two endpoints are the user (on a PC or phone) and the servers (or caches) that serve the web page.  Although we demand fast service, humans don’t really demand almost real-time response, and the heavy lifting is done on the servers within the data center.

In contrast, IoT apps' data may need to be acted on quickly (i.e., real-time sensors), and will likely use a hierarchical form of processing.  A high volume of the data is generated at the endpoints, but all of this data may not need to be all uploaded to the data center, and may be processed in intermediate stages first.

For example, sensors in an airplane engine will collect data in real time, and only a sampling or anomalies are sent to the data center for analysis, so this network will be designed as a hierarchy. Enterprises will need to design the network to offer different service levels at each point in the hierarchy. Mobile networks exhibit some of these characteristics -- spanning from a base station to a core network.

Cisco calls this fog computing, and Ericsson and others call it edge computing. In either case, processing is done at the edge so that reduces network latency.

DNS demands           

Any IoT device needs to perform DNS queries and the sheer number of IoT devices, which is estimated to be potentially in the tens of billions by 2020, will put a strain on the DNS infrastructure.  Although IoT will rely on the public DNS infrastructure, private DNS infrastructure will also get affected.

Not only are there more sources of queries, the nature of IoT devices is different.  Unlike web browsers on client end points, these devices often contain sensors where activities may occur throughout the day. DNS caches will of course assist in speeds, but it will not be enough to eliminate all queries.  What does this mean for enterprises running the apps? The data may not arrive in time (it may be queued), and you may no longer meet real-time expectations.

The traditional method for addressing this is to add more DNS servers directed by a DNS load balancer that may be behind a firewall, which becomes a bottleneck.  Vendors have proposed solutions, including F5, which sells BIG-IP DNS (formerly Global Traffic Manager) and created a solution with Infoblox that tries to address these scaling problems.

Network management

IoT devices will rely on new protocols such as AllJoyn or MQTT, which are designed for machine- to-machine communications (M2M).  This creates a challenge for the existing set of infrastructure management tools that analyze performance or packet inspection. To get ample visibility on the new protocols running on IoT networks, new network management tools for the data center and the edge may be necessary.

Managing the physical IoT devices will be a challenge due to their diversity.  Interoperability will be important, so I recommend finding some common ground to keep sane.  For large industrial devices that provide a command line interface, traditional console managers are a simple method for remote access, or a gateway can help if there are diverse connections for IoT devices (serial, USB, custom sensor connections). 

Companies such as Lantronix provide these gateway devices and have dealt with industrial devices for decades.  For other cases, it may be best to rely on a service provider such as a mobile operator or equipment vendor to create a management solution and leverage their experience.

Mobile networks

IoT devices may be anywhere and not always on wired (Ethernet) or wireless (WiFi) networks, as PCs are, so we will see a stronger reliance on mobile networks. There needs to be a platform to stitch together the enterprise with the mobile network.  The core enterprise network will share data with the IoT edge and coordinate on billing and management.

Enterprises lack deep skills for managing devices at this scale, so mobile network operators or mobile vendors can assist using their experience.

Products that can help include Ericsson’s Device Connection Platform that enables operators to manage their enterprise customers’ devices and Cisco’s Application Enablement Platform and IOx middleware.

Ultimately, I believe large IoT networks will eventually have more in common with mobile networks than traditional data center or WAN networks and the key challenge will be interoperability.


Comments

IoT

Hi Dan -- There's certainly a lot for enterprises to think about as IoT becomes more widespread, thanks for the guidance. What kind of IoT planning have you seen so far from enterprises? 

Re: IoT

Some of the IoT planning is very basic.  IP connected cameras, which are thought to be rather mundane but are actually getting rather sophisticated, can be characterized as an IoT device, and are busy delivering HD quality video upstream, all the time.  That has a effect on bandwidth usage, and also takes up another switch port that also consumes power via PoE    Many modern switches have sufficient PoE capacity, but some of them do have a limited budget that you need to be aware of.  All of that requires planning.  Given network teams are often silod, it'd be bad if someone from security says "will you give me 8 ports", and you say yes, figuring it's something harmless (maybe it's just a little badge reader?) until you realize it can affect infastructure capacity because it's actually a a set of cameras.

Another issue is putting  connected devices in branches that may have less than ideal connectivity (think of connected unmanned kiosks co-located in stores where one is a tenant of a larger store).  You are at the mercy of the other orgainztion that supplies the bandwidth to you (or more commonly will NOT give you access to their network for security reasons). In that case you do need redudant or alternate connections.

 

Re: IoT

I see, thanks for providing that insight Dan. Definitely a lot of complications!

Re: IoT

"Cisco calls this fog computing, and Ericsson and others call it edge computing. In either case, processing is done at the edge so that reduces network latency."

Dan, what is this fog computing? Is it something like grid computing or distributed computing?

Re: IoT

IT's a big different.  In fog computing - it's extending the cloud computing resources closer to the end-users -- or the edge.

In cloud computing, you have computers  in many large data centers.  But with IoT, there are so many items like sensors,  it can still be hard to connect to far away datacenters and send lots of data.

So what if you do some processing close to you and then only send some data to the cloud.  If you want to connect a smart thermostat to your own heating and cooling, you can do more processing at home or in the office.    Then you can send just a bit -- say today's high/low temp plus the # of hours the A/C was turned on to the cloud database.

 

Grid computing is differemt.  It's when you connect together many machines to run a workload together -- often in parallel.

Re: IoT

Some fog computing news this week: Cisco joined with Dell, ARM, Intel, Microsoft Corp., and the Princeton University Edge Laboratory to form the OpenFog Consortium.

Re: IoT

Interesting!  Another group that has a name that starts with the letter "O".  Or more specifically, with the word "Open".  

 

But seriously, I would like to see how Fog Computing differs from classic notions of distributed computing, or even approaches already explored in telecom.  I realize that IoT devices are probably more numerous & more widely distributed, but the question is whether one can leverage what we havem or do things really need a clean-slate.

Re: IoT

LOL! I thought the same thing -- yet another "open" industry group.

 

Re: IoT

At some point, a phrase loses its meaning.   Sun Microsystems (now part of Oracle) had a slogan "Open Systems for Open Minds", many years ago.  I think the meaning of the word open has changed, so that now it means open source as opposed to open APIs.  Actually, I really don't know what it really means :(

Re: IoT

I don't think anyone does now!

Re: IoT

"At some point, a phrase loses its meaning.   Sun Microsystems (now part of Oracle) had a slogan "Open Systems for Open Minds", many years ago.  I think the meaning of the word open has changed, so that now it means open source as opposed to open APIs.  Actually, I really don't know what it really means :("

Dan, most of the customers are also in similar dilemma. Companies may use certain attractive, stylish and polished buzz words for attracting customers; eventhough it has not any meaning with their offerings/products.  

Re: IoT

"Some fog computing news this week: Cisco joined with Dell, ARM, Intel, Microsoft Corp., and the Princeton University Edge Laboratory to form the OpenFog Consortium."

Marcia, thanks for the link. I think fog computing is also gaining momentum along with IoT and cloud.

Re: IoT

"IT's a big different.  In fog computing - it's extending the cloud computing resources closer to the end-users -- or the edge.  In cloud computing, you have computers  in many large data centers.  But with IoT, there are so many items like sensors,  it can still be hard to connect to far away datacenters and send lots of data."

Dan, thanks for the clarification and elaboration. So you meant that cloud infrastructure is placing very close to the end customers.

Re: IoT

That's correct MyNet. In fog-computing the cloud gets closer to the end-users (but there's also the regular cloud), so like fog -- it's everywhere, not just in the far-away sky.

Re: IoT

"That's correct MyNet. In fog-computing the cloud gets closer to the end-users (but there's also the regular cloud), so like fog -- it's everywhere, not just in the far-away sky."

Dan, then how it's going to differentiate from the normal private or hybrid cloud.

Re: IoT

The areas do overlap.    To differentiate, I would categorize fog computing mostly for IoT use-cases but hybrid- and private-cloud for standard workloads (i.e. web apps). But on-premises fog computing can be done by some type of cloud too - since it's just another workload.  

But you can do on-premises fog-computing with purpose systems to pre-process the data -- and they don't have to be a cloud.  A server or computer that sifts through data, and just sends a subset to a main server don't need to be designed as part fo a cloud.

Re: IoT

"The areas do overlap.    To differentiate, I would categorize fog computing mostly for IoT use-cases but hybrid- and private-cloud for standard workloads (i.e. web apps). But on-premises fog computing can be done by some type of cloud too - since it's just another workload.  "

Dan, thanks for the clarification. So it means fog computing is exclusively for IoT applications.

Re: IoT

"But you can do on-premises fog-computing with purpose systems to pre-process the data -- and they don't have to be a cloud.  A server or computer that sifts through data, and just sends a subset to a main server don't need to be designed as part fo a cloud."

Dan, if you have necessary computational; infrastructure, then there wont be a requirement for embracing cloud.  The necessity of cloud is to bridge the gap between computational/storage requirement and availability. 

Re: IoT

re: if you have necessary computational; infrastructure,  there wont be a requirement for embracing cloud. 

 

Mynet, I generally agree in the logical sense. (you may want to add "skills" to operate the infrastructure.)

But pragmatically speaking, sharing resources enables more effificient allocation of resources.  If each system is designed for peak load, then there will be lots of wasted infrstructure for off-peak period. And I'm not certain that everyone has the same skills to operate systems that a large cloud operator can do well, at scale. (one example if that small businesses use cloud-based email as opposed to running their own email servers)

So ultimately, this is an issue of distribution of resources, and that becomes a tricky economic and fairness issue too.  We all want control over resources, but we give up a bit for fairness to greater society, and there are economic benefits to using shared resources too.  Would like to understand your thoughts too.

 

 

 

Re: IoT

"I generally agree in the logical sense. (you may want to add "skills" to operate the infrastructure.). But pragmatically speaking, sharing resources enables more effificient allocation of resources.  If each system is designed for peak load, then there will be lots of wasted infrstructure for off-peak period. And I'm not certain that everyone has the same skills to operate systems that a large cloud operator can do well, at scale. (one example if that small businesses use cloud-based email as opposed to running their own email servers)"

Dan, that's true about managing the resource and skill requirements.  Am a cloud end user; but still I don't know how internally mapping or scaling happens for maximum utilization.

Re: IoT

Mynet, this may be a simple answer, but here's an example.  Resources (CPU, storage) all belong to a "pool" you can draw from.  If you have one server and one disk, the pool size is one.  You run out if there's more work or data than you can store. If you have a handful, then you can move workloads around (especially if you have something like vSphere vMotion or Live Migration, or if the apps are designed to scale-out to multi servers).  But the limit of your pool is you own data center.

But if you can extend that too using a shared set of servers on Amazon Web servies, Google Compute, or Azure or something similar AND you have ability to bring up workloads there (apps need to be modified, neworking neetds to be reworked, but it's doable) then the pool is much larger, and you can put in workloads from multiple users to share the same poool.  You can either use both your own DC plus public cloud, or just use the public cloud

So ultimately, wastage can go down IF the average utilization fits the pool size closely.  So the game is ultimately about 1) ability to expand to fit your needs and 2) make the average needs fit the capacity more closely.  And you hope that when one user's demands are low, someone else is high, so it balances out.

Of course, being too close to the limit is dangerous is there is not extra overhead to draw from.

 

 

Re: IoT

Thanks, thanks for the clarification. You simplified a complex issue to a simpler form, which is very easy to understand. Thanks for your effort. 

Re: IoT

You're welcome.  Glad to help, and I hope these discussions on Network Computing continue to help readers like you.

Re: IoT

"You're welcome.  Glad to help, and I hope these discussions on Network Computing continue to help readers like you."

Dan, yes very much and that's so kind of you