What Needs Re-imagining in the IoT Age and Why

Two industry veterans in position to offer informed perspectives on security, scalability, and the IoT’s role in an “infrastructure gold rush” share their thoughts. A look at IoT beyond the usual blather of sensors, toasters and gateways.

Editor’s note: More and more it seems the IoT looks in its closet in the morning and selects its “catalyst” outfit. With this in mind I recently lobbed a few questions to a couple of embedded computing gurus—Alan Grau, president and co-founder of Icon Labs, whose focus has been security and embedded software for more than 20 years and John Williams, vice president of marketing and product management for AppliedMicro, whose 25 years’ experience encompasses enterprise compute, storage and networking. Edited excerpts follow.

Chris (C2) Ciufo: Is the fundamental makeup of the Internet as a packet-routing mechanism about to change?

John Williams, AppliedMicro: The Internet will continue to evolve and scale as it has from IPV4 to IPV6. The physical layer and transport layers that connect the infrastructure and endpoints will also continue to evolve.

The challenge as the Internet of Things (IoT) continues to expand and the number of endpoints continues to grow exponentially is to enable mechanisms for very low-power, low-cost devices to connect to the IoT because realizing the IoT’s potential requires sensors, devices and nodes everywhere.

Figure 1: It’s the end points that enable the gathering of valuable data that will fuel innovation, says John Williams of AppliedMicro, and, when it comes to light bulbs, WiFi may not be the right choice due to cost. Photo courtesy Wikimedia Commons.

Figure 1: It’s the end points that enable the gathering of valuable data that will fuel innovation, says John Williams of AppliedMicro, and, when it comes to light bulbs, WiFi may not be the right choice due to cost. Photo courtesy Wikimedia Commons.

For example, WiFi is not necessarily the best answer for a light bulb due to its per-node cost. The ability to connect very cost-sensitive, simple devices in massive volumes is predicated on having low-cost connectivity, and the value of the data generated by these endpoints will drive innovation and new solutions.

Alan Grau, Icon Labs: The fundamental architecture for the Internet is sound and will continue to be so for some time to come. We will continue to see a rise in both mobile edge devices and “edge networks.”  That is, networks of very small devices connected together through mesh networks that are then connected to the Internet through a gateway.

C2: How is/will the IoT affect the technologies we’re discussing?

Grau: The IoT will drive the adoption of connectivity into ever-smaller devices. This will include both directly connected devices and mesh networks connected to the Internet through a gateway. End user applications including connected medical implants, fully automated manufacturing and self-driving cars will become common in the next decade. As a result, the risk of cyber-attacks will shift from data theft to cyber-physical damage with physical injury and loss-of-life—real consequences of cyber-attacks.

Williams: The IoT will create explosive growth in the number of endpoints that need to be connected and managed. It will create security challenges that likely have not even been considered in the past. Ultimately, you have significantly more endpoints generating petabytes [1015, or a million gigabytes] or zettabytes of data [1021, or a million petabytes] that needs to be transported, stored, analyzed and often archived, putting an enormous strain on data center infrastructure. Network bandwidths will need to increase in the core and at the edge of the network. Servers need to be able to store and process information much faster and at scales never previously imagined to draw conclusions and identify trends in data, often to monetize that information.

With this expansion, how the data center is architected needs to be re-imagined. Scalability of compute and storage platforms needs to be an order of magnitude better (or several orders of magnitude), and scaling is no longer a discussion of one, two, or four CPU sockets and a handful of memory DIMMs  working together in an individual server. Scalability is now a discussion of 10’s, 100’s, 1000’s or more servers working together on massive sets of data generated by the IoT.  To facilitate these new compute models, networks need to adapt from the legacy of “North-South” traffic to a new model that has a heavy “East–West” component driven by the need to efficiently scale. What that means is that data traffic is evolving – instead of traveling primarily from an endpoint to a server (North-South) we are seeing more data traveling between servers (East-West) to facilitate analysis of that data.  All of this puts enormous pressure on costs and a premium on the efficiency of solutions, resulting in unprecedented innovation in hardware and software from processors to memory to operating systems and hypervisors to data analytics software and more. The IoT is essentially triggering an infrastructure gold rush.

C2: What key hardware standards are likely to emerge as winners as the Internet evolves?

Williams: I’d like to take Ethernet for $300, Alex. But seriously, I’m still not ready to bet against Ethernet, but I will bet that what we call Ethernet today will be very different from what we call Ethernet in several years. It will emerge as a winner because it has broad support and has consistently evolved over the years.

C2: There’s a bit of an “epic battle” brewing between Intel and ARM, not in the handset but right at Intel’s bread-and-butter server business. Benchmarks show the best Xeons with the lead, but ARM has incentive to catch up. What’s your view?

Williams: Synthetic benchmarks like SpecInt Rate do favor existing x86 architectures. Unfortunately, server workloads have continued to evolve to the point where benchmarks like this are not a good representation of real world application performance. Real workload performance benchmarks often tell a very different story, and the majority of server workloads are bound by things like memory capacity, memory bandwidth, or I/O constraints like QoS.

We are at a point where compute is over-served, and what the market needs is a better balance between power efficiency, compute, memory and I/O that actually benefits application performance and Total Cost of Ownership (TCO). Just adding compute power often has little ROI versus focusing on the memory or I/O subsystem. System vendors like HP are reducing TCO with [AppliedMicro’s] X-Gene processor-based systems versus traditional x86 platforms. ARM is about innovation focused on solving the problem the most efficient way for today’s scale-out data center workloads, and these solutions are being deployed today.

C2: Experts predict video is the “killer app” to all things at the edge. What needs to change to meet the demand?

Williams: Video is a specific case of a general trend, which is the real “killer app,” of streaming data. Data from video cameras, sensors and just about everything ties very closely to the previous answer. Data compression is vital to optimizing network bandwidth utilization. In some cases lousy compression is absolutely fine, but lossless algorithms are necessary when data needs to be packaged as efficiently as possible and transported securely.

This also has implications for the other side of the process within the data center, where the ability to accelerate functions such as compression/decompression and cryptography in servers with very power-efficient, high performance offload engines is critical. Nothing is more power-efficient and higher performing than dedicated hardware. At some scale, software-based “brute-force” approaches are not the best answer even with dedicated/optimized instructions sets. The implication is that server software needs to support a variety of acceleration engines just as floating point units became integrated many years ago. Cryptography, video encode/decode and transcoding are just a few that are common enough that there is significant value in dedicating silicon to performing the function.

C2: What will “rule” in the next evolution of the Internet/network?

Grau: 6LoWPAN will play an ever increasing role as will Bluetooth Low Energy (BLE) in enabling connectivity for smaller and smaller devices. Unfortunately, security breaches are likely to continue to make headlines as many companies embrace the power of the IoT but fail to adequately address security. Standards are likely to continue to focus primarily on interoperability – with security taking a back seat in the creating of standards.

Williams: In general, data analytics software packages and software that enables the Software Defined (¬¬¬Fill in your favorite: network, server, etc.) will rule in the next evolution of the Internet /network.

C2: Give us a prediction, something you feel strongly about relating to this topic.

Williams: In the not-too-distant future, no one will ask, nor care, about what instruction set is running on a device or how many “cores” it has. Compute will be completely commoditized.

Grau: Within a decade we will read headlines of cyber-attacks resulting in loss of life, not just data breaches and stolen credit card numbers. If we don’t start addressing Internet device security now, soon there will be a major catastrophe with severe loss of life due to a cyber-security incident. Unfortunately, it may take that kind of serious incident to for the government and industry to finally develop regulations and standards to solve a problem that could have been avoided.


chris1Chris A. Ciufo is editor-in-chief for embedded content at Extension Media, which includes the EECatalog print and digital publications and website, Embedded Intel® Solutions, and other related blogs and embedded channels. He has 29 years of embedded technology experience, and has degrees in electrical engineering, and in materials science, emphasizing solid-state physics. He can be reached at cciufo@extensionmedia.com.