News

COOJA and Testbed Federation/TWIST growing together

Two COOJA plugins and manuals have been published to integrate the TWIST testbed in COOJA and to take checkpoints and perform rollbacks both in TWIST and COOJA.


Call for Papers: CONET/UBICITEC 2013

The 4th International Workshop on Networks of Cooperating Objects for Smart Cities 2013 (CONET/UBICITEC 2013), colocated with CPSWeek 2013, accepts submissions until January 28th, 2013.


Newsletter issue #19

The 19th CONET newsletter has been published. You can read on Virtual Organizations for Multi-Model Based Embedded Systems and on the UvA Bird Tracking System.


Scalable Data Processing

Partners involved

  • Cluster leader: ISEP-IPP
  • Core partners: NUIG, UCY, UDE

Introduction

Cooperating objects are able to perform functions that are difficult or impossible to perform for a single object. In particular, cooperative objects with individual sensing capabilities are able to share their sensor readings and thereby obtain a better estimate of the state of the physical environment. Typically this estimate is used to take decisions on how to actuate on the physical environment. Consider monitoring the gradient of flammable liquid within an industrial plant and taking a decision if it exceeds a certain predefined threshold. If so, then safety valves must be opened to minimize the risk of an explosion. A similar need for environmental monitoring exists in southern European countries like Cyprus and Portugal, where each year, large areas are burnt because of wildfires.

 

These types of applications require a large number of sensor nodes in order to (i) obtain an accurate image of the physical phenomena and (ii) cover the area of interest. Often, this type of applications must also be designed with what-is-available. For example, J. Elson and D. Estrin mention (in the book "Wireless Sensor Networks: A Bridge to the Physical World") a scenario where students in a class deployed a sensor network for environmental monitoring but later when a fire broke out, those sensors could be used to guide firefighters. In general, it is possible that two different sensor networks have different types of sensors that measure the same type of physical quantity (for example temperature) but one sensor network does not know that the sensor readings provided by the other sensor network are temperature readings. We conclude that scalability and the ability to deal with heterogeneity are important research issues in these types of applications.

 

Given the importance of answering queries about the physical world, the scientific community has created so-called data aggregation algorithms. Typically, sensor nodes form a tree (called routing tree or convergecast tree) with a sink node as a root. A query is broadcast to all nodes. Then leaf nodes broadcast their data. All other nodes wait until they have received a broadcast from all of their children and then make a single broadcast. Such schemes offer good performance, due to (i) the exploitation of opportunities for parallel transmission and (ii) the size of a transmitted packet may be smaller than the sum of the size of the received packets (because of in-network aggregation). But unfortunately the time-complexity of these approaches depend on the number of sensor nodes. They also suffer from the limitation that they only work inside a sensor network. These two limitations, scalability and heterogeneity, are not properly addressed by the current research literature. In fact, the need to do research on these non-functional properties "Scalability" and "Heterogeneity" has already been foreseen and emphasized in the Description-of-Work of CONET.

Before CONET started (in June 2009), individual efforts among the partners in CONET have however independently developed technologies with the potential to address and rectify the two above limitations.

  1. Researchers at ISEP in Porto have created a wireless prioritized medium access control (MAC) protocol with a very large number of priority levels. A sensor node can use its sensor reading as a priority and since the MAC protocol guarantees that the node with the highest priority wins the contention for the media and all nodes know the priority of the winner, it follows that the maximum among sensor readings can be computed with a time-complexity that is independent of the number of sensor nodes. Other quantities such as (MIN, COUNT, Interpolations) can be computed as well. This approach is promising for achieving scalability but it is currently limited to a single broadcast domain; it does not deal with heterogeneity and it cannot compute average or sum.
  2. Researchers at UCY in Nicosia have improved SQL-based query processing in networks organized as a routing/convergecast tree. An optimization is performed so that the wakeup time is small and this saves energy. An approach is also proposed for answering queries that need to compute sums where only the top-k sums need to be reported. The main idea is that an intermediate node calculates an upper and lower bound on the sum it can contribute to and it can find out whether this sum will not belong to the top-k sum and if this is the case the transmission of those tuples are suppressed. This approach is interesting because it can compute sum and average efficiently; something that the approach from ISEP is unable to do. It is also designed for networks which are not a single broadcast domain.
  3. Researchers at DERI at NUIG in Galway have created a middleware which enables the exporting of streams of sensor data from a sensor network. The middleware also supports the notion of a virtual sensor which can express an aggregated quantity in SQL. This approach is promising for (i) achieving scalability between sensor networks and (ii) dealing with heterogeneity between sensor networks. DERI's approach is relevant as it enables data aggregation across many heterogenous WSNs and also supports structured querying of those WSNs.

It can be seen that the respective partners have developed key building blocks which could potentially be combined to achieve data aggregation with good scalability and ability to deal with heterogeneity. But this potential is currently untapped. Integrative research is needed in order to advance the state-of-art.

Objective

The objective of the research cluster is to devise a solution that enables efficient and scalable query execution across multiple (large-scale) sensor networks.

Research topics

The research cluster created a reference architecture (SDP architecture) last year. The scalability that can be attained with the SDP architecture and the importance of the respective components for attaining scalability was unclear however. Therefore, the research topic of year 2 has been to create a preliminary implementation of one of the use cases of the architecture for the purpose of evaluating the SDP architecture.