· Memory footprint -- how much memory does the Platform require to function? Some simple sensors have only 128KB of memory to work with.
· Operating system support -- does the Platform require a full POSIX-like OS or can it accept something simpler?
· Network stack support, e.g. IPv4, IPv6, 6LoWPAN, other -- simple sensors used in Low Power Wireless Area Networks (LoWPAN) may require a cut down IP stack.
· Programming language support -- a Platform may provide APIs for only specific programming languages (e.g. C or C++).
· Java dependence -- does the Platform require a JVM to function, limiting sensor choices?
Murphy: The most important consideration is recognizing the risks inherent in vertically integrated solution architectures. By definition, the Internet of Things is heterogeneous in the types of things it is connecting. A horizontal architecture, to manage the information from and about the things they are connecting, can abstract the transport layer from the application layer. This allows applications to be developed independently of specific sensor devices, and sensor devices to be changed and network connectivity methods changed without breaking the application dependencies.
Schubert: A Software-Defined Machine (SDM) decouples software from the underlying hardware, making machines directly programmable through machine apps and allows connecting with virtually "any" machine and edge device, including retrofitting machines and connections to legacy systems.
* Analytics Compatibility
De La Mora: Support for structured and non-structured data, ease of integration with existing operation, automation and control systems, and the ability to operate in a distributed computing environment are all important factors for analytic compatibility.
Kester: To do advanced long-term business intelligence, machine learning or Hadoop-type of parallel processing, your Platform choice should have a well-documented and Web accessible API to interface with your analytic product of choice. It should also be easy for any IT employee, or even savvy business analysts, to use without training.
Murphy: The network platform has to enable multiple disparate audiences within a company access to benefit from data collection and perform meaningful analysis. Analytics is often thought of in a reporting sense only, but increasingly analytics is being applied in conjunction with machine learning algorithms and rules logic to drive applications and actuate devices.
Tait: You need to be sure the information you are collecting is stored well (backed up, secure, etc.) and that you have the ability to export your data and you maintain ownership.
Schubert: The tremendous data growth in industrial IoT demands massively scalable, low-cost infrastructure, such as that based on Apache Hadoop v2 and COTS (commercial off-the shelf) hardware. It has to support the various security, compliance and data privacy mandates. Predictive Analytics is how value is delivered to customers. It provides timely foresight into asset and operations, and provides actionable recommendations (when paired with rule engines). Perhaps most important, analytics need to be integrated into the operational processes, rather than be a stand-alone IT solution.
Sign up for CIO Asia eNewsletters.