Edge gateways have emerged as architectural components that improve the performance of IoT networks, and vendors have stepped up with off-the-shelf devices flexible enough to meet the varying demands of individual deployments.\nThese gateways \u2013 essentially small servers \u2013 sit between a company\u2019s data center or cloud and the IoT endpoints being used in the field.\n\nDifferent gateways will obviously be used for different tasks. A gateway that needs to perform real-time analysis of machine-generated data from a robotics-heavy factory floor will have to be at the more capable end of the spectrum, while a device that simply tracks location data in an automated fulfillment center won\u2019t need to be nearly so powerful.\n\u201cIf you\u2019re just playing around and wanted to do like a light-duty data transformation, you can get something even less capable than a Raspberry Pi for that,\u201d said FogHorn CEO David King, whose company provides a sophisticated software orchestration and intelligence layer for edge- and fog-computing setups.\nThat said, enterprise gateways are typically an off-the-rack proposition, not something custom-designed for a particular task. The biggest vendors tend to be major systems companies like HPE, Dell and Cisco. Prices for these edge devices vary based on their capabilities, form factor and level of ruggedization, but the scale reaches all the way up to potent servers capable of withstanding extremes of temperature and vibration for $2,500 and up.\nEdge gateway management\nManaging a gateway is mostly about managing the network or networks it\u2019s attached to, which means there isn\u2019t a lot of specialist knowledge or infrastructure required for supervising the day-to-day functioning of an edge gateway.\n\u201cManaging the gateway itself isn\u2019t usually a tough task,\u201d said John Spooner, a senior analyst at 451 Research. \u201cIt\u2019s managing the cable that comes off it that\u2019s typically the real chore.\u201d\nHPE's edge gateways\nThe specifics of the hardware inside any given IoT gateway are shaped by the vendor's perception of consumer requirements. HPE, for example, has seen the demand for stronger internal hardware accelerate recently.\n"Core i5 has been a sweet spot \u2013 what we\u2019ve seen is that a lot of customers start out, and they have a digital transformation they want to do or some IoT program," said Gerald Kleyn, senior director of IoT Edgeline for HPE. "So the first thing that they lack is connectivity to their industrial network, so they believe they need a gateway,"\nWhat Kleyn says many customers deploy a gateway, then find more uses for it. "[W]e have a couple gateways in our lineup, the GL10 and the GL20. A lot of those clients would install a GL10 and start collecting that data, and then immediately they\u2019d want to do more.\u201d\nKleyn said that subject-matter experts working for their clients would quickly discover that if the types of data they were collecting from industrial machines could be analyzed in real-time, it could lead to major efficiency gains across a wide range of different processes. That\u2019s often too much workload for the company\u2019s GL10 gateway, which runs an Atom processor, 4GB of RAM and a 32GB solid-state drive.\nKleyn said that more robust real-time analytics, particularly in the oil and gas industry, and robotics have been big growth areas for the more capable EL300, which boasts the aforementioned Core i5 processor.\nLarger devices, like the company\u2019s EL1000 and EL4000 units, are targeted at much the same use cases as the EL300, but simply offer higher compute capacity, enabling more complicated analytics and the ability to synthesize information coming from larger numbers of endpoints.\nDell's edge gateways\nDell, by contrast, does its best to let the proposed use case dictate the specs of the machine. Jason Shepherd, CTO of Dell's IoT solutions division, said that this can be a challenge, given the myriad use cases out there.\u00a0\n"We come up with a handful of SKUs that are purpose-built for a specific set of use cases, versus trying to build thousands of SKUs that makes it very difficult to maintain quality and scale," he said.\nTake, for example, Dell's 3000 series of edge gateways. They're all Atom-powered designs with similar core components, 2GB of RAM, and expandable internal storage, usually featuring a solid-state drive. The baseline 3001 model is designed for industrial use cases, and the differentiator is really in the type of I\/O capability it boasts. Serial connectivity, digital I\/O and having the ability to connect to both modern networked devices and old-school technologies like Fieldbus are an important part of the value proposition.\n\u201cThat\u2019s the key part of a gateway \u2013 basically, ingesting data from a bunch of different protocols and normalizing that to IP traffic,\u201d said Shepherd.\nContrast this with the 3002 model, which is designed for transportation and logistical use cases, thanks to its CAN bus interface and built-in ZigBee connectivity to help with mesh networking. It\u2019s got similar core components to the 3001, and they cost roughly the same amount at a little over $500 a pop. The difference, as Shepherd said, is in the I\/O. (The 3003 is the same type of device, equipped with DisplayPort connectivity and aimed at the digital-signage market.)\nGateway designs to suit environments\nFor all their similarities to servers in terms of internal components, edge devices usually don\u2019t look like a typical server. Thanks to the various degrees of ruggedization and different types of connectivity present in different gateways, they tend to look fairly distinctive.\n\u201cThe typical form factor of a gateway will not look like a standard, desktop-environment server \u2013 it\u2019ll be ruggedized and have a temperature, pressure, vibration spec, and it\u2019ll look like a heavy-duty device,\u201d King said, noting that it will also, generally, weigh more than an equivalent non-edge-gateway system. \u201cThe more industrialized the components and the server, the more expensive it\u2019s going to be.\u201d\nBenefits of edge architecture\nThe edge model gets used mostly as a way to keep IoT systems moving quickly. An architecture that sends data straight from the endpoint to the back-end, whether it\u2019s a cloud or a distant data center, necessarily involves delays between sending information back for processing and receiving new instructions from HQ.\nEdge gateways can handle a lot more computational work while sitting a lot closer to the endpoints, so applications with low delay tolerance and a need to have meaningful processing done quickly are best suited to an edge deployment. Robotic, precision equipment, autonomous vehicles and the like simply won\u2019t tolerate connectivity delays imposed by a full endpoint-to-cloud-and-back data cycle. Broadly, anything that has to make its own decisions in real-time is an edge candidate.\nSpooner says that IoT projects often start small, often at the ground level \u2013 something as simple as an employee attaching a monitoring device to a piece of important machinery on a factory floor. Shadow OT, if you will.\n\u201cTypically, the IT department gets involved once an IoT project gets large enough in scale or scope to where they\u2019re going company-wide, or at least large enough to start to ask for resources to help support their project,\u201d he said.