This post is an extract of the articles “Integrating Mobile Internet of Things and Cloud Computing towards Scalability: Lessons Learned from Existing Fog Computing Architectures and Solutions” presented at 3rd International IBM Cloud Academy Conference (ICACON) in Budapest (Hungary) in 2015 and published at International Journal of Cloud Computing.

Fog computing [1] is a recently introduced and already popular term, coined by Cisco, to indicate the possibility of moving (usually virtualized) computing elements close to the physical position of end-users, especially of mobile devices in the IoT. Fog is considered a driver for enterprise/industrial-based MIoT that brings connectivity to the real world in a way never reached before and tries to face new business models introduced by the disruptive IoT concept. In particular, fog computing is recognized as a paradigm that can relevantly enhance the efficient applicability of cloud computing techniques to the MIoT, enabling edge computing gateways to smoothen the potential efficiency mismatch coming from a straightforward “dumb” integration between IoT devices and the cloud, e.g.:

  • high frequency of geographically distributed interactions
  • all control and management intelligence virtualized only on the global cloud resources, no scalable decentralization over localities.

In fact, in this perspective, fog computing can be seen as a distributed, resource-limited, first-level cloud that receives MIoT data and performs (partial) analysis of them immediately and locally, by enabling localized, decentralized, and prompt control actions when needed. This is particularly crucial when the addressed MIoT scenarios include actuation devices, as in cyber-physical systems. If the “localized intelligence” of fog computing cannot autonomously complete the application-level business logic, as usual, anyway there is the advantage of sending only pre-processed data (possibly batched) to global cloud resources, e.g., for resource-intensive processing, storage, or long-term analysis.

Future envisioned MIoT applications are large–scale and latency–sensitive ones, no longer created to work alone, but to share infrastructure, communication resources and common management platforms. Those applications require new specifications to be satisfied, like mobility support, large-scale geographic distribution, context awareness, low latency, and low traffic overhead in order to meet quality requirements, such as scalability. Due to global availability, geographic distribution, and the best-effort Internet, cloud computing resources alone and by themselves cannot satisfy those new requirements and critical issues have to be faced to turn new MIoT applications suitable to be deployed in real-world scenarios. Hence, cloud remains central with its unbeatable efficiencies in terms of elasticity, scale and cost-savings, but the complementarity and interplay between fog elements are key to achieve the purpose.

The fog computing vision was conceived to address some of the related technical challenges [2]: several research works have already started to recognize the issues associated with direct communication between MIoT devices and globally available cloud resources. [3, 4] propose new architectures that will allow the integration of lightweight sensors with the cloud, by overcoming issues such as latency, management of continuous sensing, ability to support periodic events, and lack of elasticity when numerous wireless sensors transmit data simultaneously. [5] specifically addresses the problems deriving from continuous sensing when straightforwardly integrated with the cloud, such as energy consumption and communication overhead/cost. [6] focuses on network latency impact: some experiments are reported to show that generally the most sensitive latency is from locality edges to the cloud and this is particularly true for cases where edges need to retrieve large amounts of data and have strict response time requirements.

For further details, please also refer to this presentation where you can find the state-of-the-art in the field, its relevance for future large-scale MIoT applications, existing architectures will be rapidly discussed and used as the basis for the proposal of an original architecture and taxonomy. In addition, the paper will report lessons learned from the few existing MIoT-cloud integration experiences and, based on this overview and original architecture/taxonomy proposal, we will highlight specific sub-fields of future research where advancements are needed.

  1. F. Bonomi, R. Milito, J. Zhu, S. Addepalli, “Fog Computing and Its Role in the Internet of Things”, Int. Workshop on Mobile Cloud Computing (MCC), 2012.
  2. F. Bonomi, R. Milito, P. Natarajan, J. Zhu, “Fog Computing: A Platform for Internet of Things and Analytics”, book chapter in “Big Data and Internet of Things: A Roadmap for Smart Environments”, Springer, 2014.
  3. W. Wang, K. Lee, D. Murray, “Integrating Sensors with the Cloud using Dynamic Proxies”, IEEE Int. Symp. Personal Indoor and Mobile Radio Communications(PIMRC), 2012.
  4. P. Zarko, A. Antonic, K. Pripuzic, “Publish/Subscribe Middleware for Energy-Efficient Mobile Crowdsensing”, ACM Conf. Pervasive and Ubiquitous Computing (UbiComp), 2013.
  5. N.D. Lane, E. Miluzzo, H. Lu, D. Peebles, T. Choudhury, A.T. Campbell, “A Survey of Mobile Phone Sensing”, IEEE Communications Magazine, Vol. 48, No. 9, pp. 140-150, Sept. 2010.
  6. A. Faisal, D. Petriu, M. Woodside, “Network Latency Impact on Performance of Software Deployed Across Multiple Clouds”, Int. Conf. Center for Advanced Studies on Collaborative Research (CASCON), 2013.