Presentation of the collaboration



Program and results

We recall that the MINES work-program for the 3 years is structured according to the following themes:
  • T1 - Adaptive communication architecture for resilient sensing and actuation.
  • T2 - Stochastic models for analysis and runtime decision.
  • T3 - Cross-layer in-network collaboration protocols to enhance the overall QoS.
  • T4 - Middleware solution that embeds the models and implements the adaptive protocols.
  • T5 - Evaluation using focused emergency scenario use cases.
As planned, our Y1 focus was on the first 3 themes; we summarize below the related contributions.

2018

T1 - Adaptive communication architecture for resilient sensing and actuation

As we move further into a future full of connected devices, the Internet of Things (IoT) promises to revolutionize societal-scale operations and influence our daily lives. It integrates pervasive sensing/actuation, dynamic data analytics, and communications. Domains such as transportation, home automation, healthcare, and emergency response are becoming increasingly IoT-enabled; this provides data-driven insights to improve situational awareness. This is particularly useful in mission critical applications, e.g., to enable effective and timely emergency response. Recent smart city efforts such as the Smart America Challenge and Global City Teams Challenge have showcased the integration of IoT into a variety of community settings and application domains.

A distributed data exchange solution that manages the flow of relevant data to/from devices, systems and individuals (data producers and consumers) is a critical centerpiece of IoT deployments. In our work, we adopt a publish/subscribe (pub/sub) model for IoT data exchange based on our previous experiences with such systems and the popular use of pub/sub in IoT implementations.

Following, to enable resilient sensing and actuation, we designed the architecture of FireDeX by relying on a cross-layer approach:

  • Application layer: FireDeX peers (e.g., firefighters) subscribe their interest for receiving data from the IoT enabled building. Because different data vary by importance, we propose prioritizing events according to their relative importance to the emergency response effort. To configure this, subscribers register utility functions with their subscriptions. These functions capture a quantified measure of value for varying rates of event delivery performance.
  • Data exchange layer: At this layer, our proposed algorithms prioritize subscriptions according to their subscriber-specified utility functions. This is enabled by leveraging a queueing theoretic analytical model that estimates system performance under a given configuration. This is given as input to our algorithms that assign discrete priority classes to allocate available network bandwidth.
  • Network layer: This layer manages the network infrastructure through APIs provided by an SDN controller that likely runs alongside the BMS (i.e., at the edge). To enforce event priorities at the network layer, FireDeX leverages SDN APIs. It configures priority queueing disciplines for packets matching the different subscriptions. However, for the network to distinguish the data exchange-layer concept of subscriptions, we first translate it to a network-level concept.

The above architecture enables system designers to prioritizing mission critical IoT data exchange during an emergency scenario in an IoT-enhanced smart building with SDN-enabled infrastructure regardless of the message broker used.

T2 - Stochastic models for analysis and runtime decision

By relying on the above architecture, we have formulated a generalized model for prioritized data exchange in mission-critical settings. Hence, we must consider all three layers' characteristics and their effects on each other. However, existing efforts typically focus on each layer in isolation. Accordingly, we have modeled cross-layer interactions by composing and extending previous work at each layer through the unified framework of queueing theory. This model includes publication rates and subscriptions at the application layer. Simple M/M/1 or G/G/1 queues are used for representing subscription matching at the data exchange layer, as well as our new multi-class priority queueing model at the network infrastructure. The latter queueing model is used to represent an SDN switch, but it is generally suitable for use in other queueing networks. This analysis results into an analytical model for estimating a particular configuration's expected performance. Subsequently, it is leveraged as input to our algorithms that prioritize IoT events and tune notification delivery/delay.

T3 - Cross-layer in-network collaboration protocols

In order to increase the reliability and accuracy of small, inexpensive things, our plan within the theme is to adapt and extend the team’s previous works on crowdsourced mobile data and calibration together to improve the quality of the crowd-sensed data. Still, to ease the combination of the two lines of work, part of our 2018 work has been focused on consolidating the individual pieces of work, along the following lines:

Collaborative calibration for mobile-crowdsensing scenarios: To monitor the urban environment, large cities typically rely on few sensing stations that provide highly accurate measurements. Nevertheless, few stations are not sufficient to provide measurements with high spatio-temporal resolution and representativeness. To tackle this issue, cities tend to take advantage of the IoT and especially of the proliferation of mobile devices with increased sensing capabilities. This has given rise to the development of the crowd-sensing movement wherein users (citizens, groups, communities) are engaged into some collaborative data collection, analysis and decision making. In practice, citizens rely on the small and low-cost sensors embedded or connected to their smart-phones so as to help identifying environmental problems. Still, crowd-sensing the urban space raises a key challenge, which is mainly related to data quality: cheap sensors are subject to noise, bias and drift that render questionable the quality of the acquired data. To tackle this issue, we study macro calibration systems that compensate the sensor errors while alleviating the need for visiting iteratively each sensor to manually (re-)calibrate. In particular, we study distributed calibration solutions that opportunistically leverage the presence of the nearby sensors, which monitor the same phenomenon.

In our previous work, we introduced the mathematical framework for calibrating mobile sensors, and briefly sketched a mathematical formulation. We have been extending this work with a detailed description of our model and the algorithms used to calibrate, in addition to addressing the case where a selection of the best device(s) to calibrate is supported.

User-Centric context inference for mobile crowdsensing scenarios: Mobile crowdsensing is a powerful mechanism to aggregate hyperlocal knowledge about the environment. Users may indeed contribute valuable observations across time and space using the sensors embedded in their smartphones. However, the relevance of the provided measurements depends on the adequacy of the sensing context with respect to the phenomena that are analyzed. This paper concentrates more specifically on assessing the sensing context when gathering observations about the physical environment for which one needs to characterize the location of the smartphone beyond its geographical position in the Euclidean space, i.e., whether the phone is in-/out-pocket, in-/out-door and on-/under-ground. While the literature documents approaches that focus on characterizing one of these three context information, we tackle the inference of the mobile sensing context as a whole. We specifically introduce an online learning approach to the local inference of the mobile sensing context so as to overcome the disparity of the classification performance due to the heterogeneity of the sensing devices as well as the diversity of user behavior and usage scenarios. Our approach further features a hierarchical and adaptive inference algorithm that requires few opportunistic feedbacks from the user, while increasing the accuracy of the context inference per user.

Calibration of the infrastructure sensors: The recent advances in low cost, low power and multi-functional sensors, that are miniaturized and networked has led to their increased usage to deal with tactical and emergency response. Sensors are thrown in mass so as to safely sense various phenomena of interest, e.g., the temperature, the concentration of toxic material in the air, the sound generated by rescued persons, and, henceforth to monitor the overall evolution of the disaster (e.g. a fire). However, mission-critical applications should address additional requirements related to robustness and reliability. In particular, the preparation begins long before the actual incident: operational sensors cannot be left unattended for a (long) period of time, without compromising their reliability and availability. To tackle this issue, we propose that a team (re-)calibrate and (whenever necessary) replaces some of the sensors that have been massively deployed. Rather than manually calibrating sensors in a laboratory (as it is traditionally the case), we equip a team with several high-quality sensors that are freshly calibrated and serve as references. When a team member is near a set of uncalibrated sensors, then the calibration takes place. Going one step further, we attempt to coordinate the team members to calibrate many sensors.

In a conjoint effort and building upon our respective background, we are exploring opportunities to plan the maintenance and calibration of the sensors that have been deployed in a smart building with regards to a community scenario. We are formulating a two-level planning problem -- on the higher level, we determine which sensors to be visited for maintenance purpose; on the lower level, we plan the paths for the mobile team members to actually visit their deployment sites. The overall objective is to lower the average cost (e.g., time) in recurring jobs that span over a long period of time. We are now developing algorithms to solve this optimization and extending our existing platform to support experiments.

2019

Building upon the 2018 results, the 2019 work program will address the 5 MINES themes as follows.

T1 - Adaptive communication architecture for resilient sensing and actuation

As part of T1 we have implemented the FireDeX middleware based on the presented cross-layer architecture. During the 2nd year’s program we plan to develop an experimental framework that simulates the physical network of FireDeX with Mininet. Then, we will rely on our real-testbed at UC Irvine to feed the middleware with streams of data. Peers interested in such data feeds (i.e., firefighters) will be imported through configuration files. The main purpose of these experiments, will be to calculate end-to-end response times and delivery success rates in order to validate the analytical models of 1st year’s T2.

T2 - Stochastic models for analysis and runtime decision

The analytical and simulation models identified during the 1st year, can be mainly used for design-time system tuning in order to ensure accurate runtime system behavior. However, emergency scenarios are mainly characterized from dynamic conditions and thus we intend to study aspects such as:

  • Failing publisher devices – i.e., camera devices may brake because of the fire.
  • Subscriber churn – i.e., subscriptions may be discontinued.
  • Varying network bandwidth/error rates.
  • Changing utility functions – i.e., firefighters may change their preferences during the operation.

To deal with the above aspects, we plan to rely on the identified queuing models and several other techniques, such as reinforcement learning in order to re-estimate publication rates, adapt the system to varying bandwidth and re-assign event priorities and drop rates.

T3 - Cross-layer in-network collaboration protocols to enhance the overall QoS

Complementary to the above studies, T3 is concerned with the study of coordination protocols across Things so that they collaborate toward the goal of the system they contribute to, while both enhancing the quality of the delivered data and optimizing the consumption of the network resource.

In particular, our objective within T3 for the coming year is twofold: (i) Finalizing the design and development of a coordination framework aimed at the maintenance and calibration of sensors that are located in a smart environment; (ii) Bearing in mind the emergency-response scenarios that MINES targets, we will continue investigating protocols that contribute to enhancing the quality of the observations through context-aware collaborative sensing and associated macro-calibration.

T4 - MINES middleware solution

The T4 theme is focused on the study of the overall MINES middleware solution for emergency scenarios, which in particular build upon the contributions under the T1 to T3 themes.

Following, during Y2, we will study the design of the overall middleware architecture, which integrates the components/protocols for an adaptive communication targeting resilient sensing and actuation and integrating a stochastic model for runtime analysis, as developed within T1 and T2, with the engine for supporting a cross-layer in-network collaboration and calibration as developed within T3.

T5 - Evaluation using focused emergency scenario use cases

The research work within T4 is primarily intended for the 3rd year. Still, within the second year, we will start elaborating specific use cases, which will serve evaluating the MINES middleware solutions in the course of the 3rd year.

Top