Guest Editor’s Introduction • Arkady Zaslavsky • October 2015
Translations by Osvaldo Perez and Tiejun Huang
Listen to the Guest Editors’ Introduction
Analytics is generally defined as the discovery and multimodal communication of meaningful patterns in data that can be stored locally or in the cloud. Mobile analytics emphasizes lightweight discovery on mobile devices, computation balancing, and communication of meaningful patterns in sensed, observed, discovered, and computed data while on the move. Mobile analytics is concerned with data streams and query processing on mobile devices (such as smartphones) and adaptive distributed cooperation with cloud/fog-based resources. Cisco recently introduced fog computing (http://www.cisco.com/web/solutions/trends/iot/fog-computing.html ) to emphasize distributed processing of Internet of Things (IoT) data including on-edge devices such as gateways, routers, and smartphones. Analytics includes the offline as well as online processing of data — potentially big data. Mobile data analytics comprises a set of tools that process, analyze, and visualize data originating from mobile devices and other sources (for example, social media and wireless sensor networks) and consumed by mobile devices while on the move.
The past decade has seen the emergence of smartphones equipped with powerful computational processing capabilities that include a range of sensors and Internet connectivity capabilities through wireless communication networks. Technology has transformed mobile phones from mere voice telephony devices to rich data sources. The data from these ubiquitous devices offers exciting opportunities for the development of novel IoT applications. Mobile devices equipped with sensors and communication capabilities can act as a bridge to embedded IoT objects (coffee machines, fridges, industrial robots, and so on), or generate information about the environment (for example, air-quality measurements). In many cases, an app transmits the data that the mobile devices generate to a remote server/cloud for further processing. The increasing amounts of data that mobile devices generate and consume has created significant research interest in mobile data analytics in recent years.
This month’s Computing Now theme begins with “ShareLikesCrowd: Mobile Analytics for Participatory Sensing and Crowd-sourcing Applications” by Arkady Zaslavsky, Prem Prakash Jayaraman, and Shonali Krishnaswamy. The authors introduce a taxonomy of mobile analytics systems and analytics on data that can be performed both locally (mobile device) and remotely (server/cloud/fog). Figure 1 presents a broad classification of mobile analytics enabled systems which in one way or another depend on sensor data originating from mobile devices. The sense component in Figure 1 could include data originating from mobile devices’ onboard sensors, user response to questions in crowd-sourcing applications, and social networking data originating from mobile users. The article classifies applications that employ mobile analytics into four categories:
- push/pull-based independent systems with local sensing, cloud processing and with/without local processing;
- push-based collaborated systems with local sensing and processing;
- push-based collaborated systems with local sensing and cloud processing/storage;
- and push/pull-based collaborated system with distributed processing and load balancing between cloud and mobile device.
The most prominent local analysis is conversion of raw sensed values into application usable values (for example, an analog-to-digital convertor). A key challenge for mobile analytics is whether to perform data analysis locally or in the cloud. Various crowd-sensing and sourcing reference architectures employ different combinations — that is, local, cloud, or a combination of both. Another key challenge is determining the heuristics and algorithms that achieve the desired application functionality. Energy-efficiency is also a critical factor in deciding the load-balancing strategy. These heuristics can range from data-mediation techniques such as filtering and noise elimination to context inference (such as user activity and noise levels), and data mining approaches, such as clustering, change detection, decision trees and others. The article also introduces a data stream mining system called CAROMM (Context-Aware Open Mobile Mining), a mobile crowd-sensing framework that uses a scalable data collection engine to deliver real-time and situation-aware location information for mobile users while offering mobile analytics processing capabilities.
In “On the Link(s) Between ‘D’ and ‘A’ in Mobile Data Analytics,” Goce Trajcevski explores efficiency management for applications relying on mobile data analytics. He focuses on spatio-temporal data related to mobile objects and argues that gathering, storing, retrieving and querying spatio-temporal data separately from executing control and decision-making algorithms might not be beneficial. He argues that an important component of mobile analytics is the management of the dynamics associated with different trade-offs and, more importantly, linking them when orchestrating the data management and analytics processes. The article presents the DNA2 (Data’s Natural Associations with Analytics) hypothesis for coupling data with control. Trajcevski details the role of uncertainty and aspects of declarative specifications for merging reactive and proactive behaviors in applications that relyg on mobile analytics.
Swarnava Dey and his colleagues’ “Challenges of Using Edge Devices in IoT Computation Grids” proposes an architecture in which smartphones can act as edge gateways between sensor networks and cloud platforms. They argue that the increased use of smartphones’ computational capabilities must match the amount of data the IoT is generating. He discusses the challenges of creating such an ecosystem and proposes a scheme in which smartphones, residential gateways, and other edge devices can balance the computational load efficiently and effectively.
Figure 1. Broad Classification of Mobile Analytics Enabled Systems
Aleksandr Antonić and his colleagues propose an ecosystem for mobile crowd-sensing applications that relies on Cloud-based Publish/Subscribe middleware (CUPUS) to acquire sensor data from mobile devices in a context-aware and energy-efficient manner in “A Mobile Crowdsensing Ecosystem Enabled by a Cloud-based Publish/Subscribe Middleware.” The proposed ecosystem offers the means for location management of mobile IoT objects and adaptive data acquisition from such devices. Antonić and colleagues’ solution enables the filtering of sensor data on mobile devices in the proximity of a data producer prior to its transmission into the cloud. Thus it reduces both network traffic and energy consumption on mobile devices. The authors evaluated the performance of a mobile CUPUS application to investigate its performance on smartphones in terms of scalability and CPU, memory, and energy consumption under high publishing load. The developed CUPUS code is open source as part of an IoT middleware OpenIoT platform.
“Edge Analytics in the Internet of Things,” by Mahadev Satyanarayanan and his colleagues, describes GigaSight–an Internet-scale repository of privacy-preserving crowd-sourced video content. The architecture represents a federated system of VM-based cloudlets that perform video analytics at the edge of the Internet. The cloudlets can run on smartphones or, more generally, on any mobile device, and the processing location is based on cost-efficiency criteria. Privacy-preservation impacts how the system modifies video streams and what information it sends to the cloud. The GigaSight architecture focuses on intelligent transportation systems in which vehicle-mounted cameras generate enormous amounts of data that could congest the network in the absence of smart analytics.
Ahmad Razip and his colleagues explore the design and use of a mobile visual analytics toolkit for public safety data that equips law enforcement agencies with effective situation awareness and risk assessment tools in “A Mobile Visual Analytics Approach for Law Enforcement Situation Awareness.” The proposed system provides users with a suite of interactive tools that let them perform analysis and detect trends, patterns, and anomalies among criminal, traffic, and civil incidents. It also provides interactive risk assessment tools that let users identify regions of potential high risk and determine the risk at any user-specified location and time. The system has been designed for the iPhone/iPad environment, and the authors indicate that a consortium of law enforcement agencies is now evaluating the system.
This month’s theme presents two video interviews of prominent industry experts working in mobile analytics. I asked these experts to comment on the following questions:
- How would you define mobile analytics?
- Where do you see mobile analytics R&D going in the next few years?
- Can you give examples of mobile analytics projects in your organisation and the impact these projects make?
The first video features, Amy Shi-Nash, chief data scientist for DataSpark Pte Ltd, which is associated with Singapore Telecommunications (Singtel) Ltd. The second video presents a mobile analytics perspective from Anirban Mondal, a researcher with Xerox Research Centre India (XRCI).
These articles address various important challenges that mobile analytics raises and propose solutions that will make mobile analytics — as one component of the Internet of Things — practicable, feasible, deployable, and usable. Of course, more research and many more challenges should be addressed and resolved before mobile analytics becomes a truly useful enabler and a feature of our smartphones in the next five to seven years. Interested readers are encouraged to conduct further research and join the community of practice for mobile analytics champions, researchers, developers, architects, and users.
Arkady Zaslavsky is senior principal research scientist at Data61 of the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia. He is also an adjunct professor at the University of New South Wales (UNSW), La Trobe University (Melbourne), and Luleå University of Technology, Sweden. He has a PhD in Computer Science from the Institute for Control Sciences (IPU-IAT), USSR Academy of Sciences. His technical interests focus on the Internet of Things, pervasive, ubiquitous, mobile computing, context-awareness, semantic data management and mobile analytics. Zaslavsky is a member of the CN editorial board. Contact him at firstname.lastname@example.org.