Since he announced it in 2011, Marc Andreessen\u2019s statement that \u201csoftware is eating the world\u201d has become a truism with a pride of place in every presenter\u2019s deck (full disclosure: it\u2019s featured prominently in mine).\u00a0\nRecently, Benedict Evans of the venture capital firm Andreessen Horowitz (seeing a pattern here?) writes that \u201cmobile is eating the world.\u201d He asserts that mobile phones will achieve the following:\n\nBecome the de facto Internet access device.\nBring billions of new users onto the Internet\u00a0\n\nWithout disagreeing with either of these statements (actually, I agree strongly with both), their statements reflect points in time. The future of our society and economy is digital, yes. Humans will overwhelmingly use mobile devices to carry on their Internet-based daily lives.\u00a0\nHowever, both of these statements overlook one salient fact of technology history: the inexorable trend of computing devices, due to Moore\u2019s Law, is that useful work can be done in smaller and smaller form factors, which, due to greater convenience and lower prices, explode in deployed volumes. The unstoppable system architecture of the future is a small device talking to a cloud-based application. The question is, what is that small device? I assert that mobile phones will come to represent a tiny percentage of the client devices operating on the Internet. Why?\u00a0\nBecause located out beyond the mobile phone horizon is the next form factor: the Internet of Things. These are even smaller computing devices that are typically devoted to performing one function, unlike mobile phones, which are the Swiss Army Knives of computing, serving as personal transportation coordinators, activity trackers, movie ticket buyers, and a myriad of other tasks.\u00a0\nAnother reality of IoT devices is that there will be hundreds of billions of them, vastly outstripping even the number of mobile phones in the hands of users. Moreover, most of them will not be used by humans, but will be standalone devices monitoring some environmental factor or effectuating some physical action (referred to as sensors and actuators, respectively).\u00a0\n[Related: The Internet of Things comes to the NFL]\u00a0\nWhile most people are aware of IoT, I feel that they significantly underestimate what IoT will represent. IoT will be much more omnipresent and highly varied that most of us can understand. Literally, every physical thing in our environment is likely to be remade as a connected device.\u00a0\nThis variety was brought home to me last week at Cloud Expo Asia (CEA), held in Singapore. CEA is a fascinating show, held in a fascinating place. I always enjoy spending time in Singapore because it is the crossroads of Asia and is a nation with enormous energy and ambition, located in a region that, in many ways, represents the leading edge of trends that will eventually wash up on the shores of Europe and the America.\u00a0\nAt CEA, I saw two presentations that exemplify powerful examples of the power of IoT as well as the wildly different kinds of solutions that reside underneath the umbrella of the term.\u00a0\nFirst up was a presentation by the global BI head of Tetra Pak, Fredrik Ohlsson. Tetra Pak, if you\u2019re not familiar with it, creates containers. You may have seen these holding milk \u2013 they look like waxy cardboard, but, due to their design and fill system, can be stored at room temperature. Here is a video that explain how Tetra Pak works \u2013 it\u2019s pretty amazing, as it sort of forms the package around the product as it\u2019s filled.\u00a0\nTetra Pak is a global company and faces many challenges in data consistency, summarizing data, allowing locally-focused analytics, and so on. Ohlsson\u2019s presentation addressed how Tetra Pak has improved its business intelligence approach over the past few years to provide better business insights. An excellent initiative and presentation.\u00a0\nWhat struck me was a response he gave to a question raised at the end of his presentation \u2013 could he provide some examples of business advantage Tetra Pak had achieved via its updated BI capability. He responded that Tetra Pak treated this as privileged information, but he could offer one example.\u00a0\nAs demonstrated in the video, Tetra Pak milk carton machines cut the packaging material to form individual milk containers. Occasionally, the machine\u2019s blades get dull, and instead of cutting a container, mash the packaging material, allowing milk to spill all over the area. Recovering from a spill takes time, and, worse, holds up production.\u00a0\nHe noted that they are now capturing cutting data from the filling machines and see that, before a blade gets dull enough to mash the packaging material, its cuts take slightly longer than when the blade was newly installed. By tracking the cutting time, customers can see when a blade needs to be replaced, which can be done at a convenient time that does not halt production. No mess, no downtime, money saved.\u00a0\nA second presentation by Gul Agha discussed a very different IoT application. Gul teaches at the University of Illinois, Urbana-Champaign, and his work focuses on smart cities. The example he discussed is bridge safety monitoring. Bridges have to be monitored for corrosion, element failure, and so on. The consequences of monitoring failure can be catastrophic.\u00a0\nWhile in this particular example, human monitoring had identified serious issues, many bridges pose significant problems for monitoring: remote location, difficult access to key structure elements, infrequent monitoring schedules due to location, etc.\u00a0\nAgha outlined a solution his team has developed. Small sensor devices are placed on the bridge and gather data around the clock. With the approach, potential issues can be identified much earlier; moreover, this solution mitigates the location\/element challenges just mentioned.\u00a0\nHowever, the convenience of sensor location still faces other issues: how to power the sensors and associated electronics containing the computing equipment, network connectivity to forward information to a centralized data repository, and how to communicate data in severely bandwidth-constrained environments. This latter issue can constrain the ability of the sensors to directly forward data readings to the repository.\u00a0\nAgha described some of the approaches to address these issues:\u00a0\n\nIn many locations, solar cells can provide limited amounts of power to the sensor devices. However, solar cell size restrictions mean that only a limited amount of power can be supplied.\nNetwork connectivity can sometimes be achieved via cellular means. When the location is too remote for cellular communication, sensors can use mesh networking to forward data to a location that can make a network connection. If neither of these approaches can work, a vehicle (train or truck) can carry a receiver, which gathers the data from the devices, which amusingly reminded me of this old method of picking up mail via trains.\nTo address constrained bandwidth, the sensors discard individual data readings in favor of summaries. This enables a once-a-day aggregated data to be forwarded to the central repository.\nFinally, the individual sensor devices have to be smarter than mere readers of data and capable of operating as independent actors that can operate with more autonomy. Dr. Agha is an expert in actor-based programming, which is an emerging trend in languages and is used by both Facebook and Microsoft.\u00a0\n\n[Related: How the Internet of Things is changing healthcare and transportation]\u00a0\nWhat these two examples brought home is how the term IoT shields the enormous complexity associated with individual sensor\/actuator solutions. Four key elements are associated with every IoT solution:\u00a0\n\nPower: Tetra Pak presumably had plenty of mains power for its \u201cdevice\u201d (the milk container filling machine). The bridge sensors typically have access to very limited amounts of power.\nConnectivity: The type, availability, and bandwidth size can vary significantly, depending upon where the IoT device is placed.\nPayload size: Some solutions may be forced to drop individual data records and rely on summaries, while others may be able to capture and communicate large amounts of data.\nComputing capability: Some solutions may find \u201cdumb\u201d sensors\/actuators sufficient for system functionality. Others may require more capable devices that can operate more autonomously.\u00a0\n\nEach IoT solution needs to identify system resources and constraints and use them as solution design inputs. In a way, this reminds me of the early days of the Internet, in which websites had to be designed with highly restrictive user bandwidth limitations in mind. Boo.com was an infamous example of a website whose design outstripped the capabilities of user compute and network resources. This led, as you might imagine, to an unfortunate end for the company.\u00a0\nIt\u2019s likely we\u2019ll see some of the same failed solutions in the IoT space. However, just as Boo.com\u2019s failure did not consign ecommerce to history\u2019s dustbin (Amazon seems to be doing fairly well), just so will IoT failures come to be seen as individual disasters. Long-term, IoT will emerge as a gigantic use case that comprises the vast preponderance of overall computing activities.