OVERVIEW

Sensor networks will consist of a large collection of small nodes providing collaborative and distributed sensing ability in unpredictable environments. Given their unattended nature, it is important to have an infrastructure to provide indication of node failures, resource depletion, and other abnormality.    Such information can provide early warning of system failure, and guidance for incremental deployment. It can also serve to self-test for known external stimuli, or discover unknown stimuli.

In general, the key challenge in monitoring sensor networks is to collect network health indications from within the sensor network in a manner that scales well with network size, is robust to node failures, and is energy efficient. The last constraint, in particular, cannot be over emphasized: Monitoring activity is inherently energy consuming, and we need to be especially careful in the design of mechanisms for collecting monitoring data.

APPROACHes

In this project, we propose an architecture for monitoring sensor networks consisting of three components: Crucial system performance metrics are collected by network aggregate computation in the background. Occasionally, scans are invoked to provide global views of system state. In the debugging phase to isolate the problem, detailed node states or logs are collected by dump. This infrastructure provides users flexibility to extract system status with different levels of detail while alleviating the overhead of monitoring activities.

This project proposes an infrastructure to monitor and manage wireless sensor networks with different levels of detail while we alleviate the overhead of monitoring activities. We categorize three different types of monitoring tools that are complementary to each other:

The first component consists of tools such as dump. Upon a user's request, dump collects detailed node states or logs over the network for diagnosis. For example, we could dump the raw temperature readings from some sensors to debug the collaborative event detection algorithm between nearby nodes. Another example is LinkScan, which collects network topology information (neighbour list) from the targeted region. LinkScan can be implemented as an application upon directed diffusion. Because the amount of data per node may be large, such tools should only be invoked for a small fraction of the network at any one time.

Global views of node states are collected by another set of tools similar to escan. In escan, a special user-gateway node initiates collection of node state (for instance residual energy supply level) from every node in the system. Instead of delivering the raw data to user node, escan system takes advantage of in-network aggregation: Residual energy level data from individual nodes are combined into more compact forms, if and only if those nodes are nearby and have similar energy level. By pushing the data processing into the network, escan constructs an approximate system-wide view of energy supply levels with much less communication cost compared to centralized collection. From such a global view, users are able to isolate those nodes on which to start dump. Inherently, escan attempts to collect information from everywhere in the network. The incurred cost enfeebles the possibility of running escan continuously for large networks.

Finally, we propose an early warning system by continuously computing aggregate network properties on the background. Each aggregate property consists of only a few bytes of data. It denotes some critical metrics of system health. For example, the size of network i.e. the number of nodes, can indicate several system health conditions: Sudden drop in the network size can be taken as a hint for massive node failure or network partitioning. While oscillation in reported network size may result from common existence of intermittent links. Another example is the average degree of the network that quantifies how well the nodes are connected in general. A few bytes of information is hardly sufficient to isolate the underlying problem but is able to alert users to invoke appropriate tools such as scan and dump for further investigation.

ACCOMPLISHMENTS

Motivated by difficulties we encountered in designing and implementing monitoring tools on real systems, we conducted a systematic measurement study of packet loss on RFM based motes. With approximately 60 MICA motes equipped with RFM radio, we studied packet delivery performance in three different environments (In-door, Obstructed-Parking-Lot and State-Park). We evaluated packet loss in those environments and its correlation to distance, signal strength and transmission power. We also carefully studied temporal variance in packet loss and asymmetry in packet delivery. This study quantitatively revealed the dynamic range of packet delivery in various environments.

One of the primary results from our study is the spatial characteristics of packet delivery. The Figure below on the left plots the spatial reception profile of receivers at different distance for an indoor environment, where a transmitter was placed at distance 0 transmitting at maximum transmit power with 4b6b coding we notice a very interesting phenomenon: There are two distinct regimes of reception rate: up to a certain distance from the sender, packet reception rates are uniformly high. Beyond this, however, there exists a gray area in which reception rate varies dramatically; some nodes see near 90% successful reception, while neighboring nodes sometimes see less than 50% reception rate.

More interesting than the existence of the gray area is the extent of the gray area. The width of the gray area is almost one third the total communication range.  For the habitat environment (Plot omitted), the gray area covers one fifth. While it has been known (at least anecdotally) that nodes at the “edge" of the communication range often see erratic packet reception, this “edge" is significantly thick in some environments.  The width of the gray area has fairly deep implications for sensor networks. Sensor networks are designed to provide fine-grain monitoring of physical phenomena, which implies dense deployment in possibly harsh environments. The existence of a gray area implies that the likelihood of links falling into the gray area is high. For example, consider a habitat environment where the width of the gray area is 1/5th of the communication range. If we assume uniform deployment of sensor nodes around a given node, it follows that about 9/25th (or nearly a third) of a node's neighbors are likely to be in the gray area!

Not only do the links in the grey area suffer choppy reception, they also show high dynamics in packet delivery. The figure above on the right describes the computed standard deviation of average reception rate over a 40 sec time window, computed from data taken over 2 hours for each receiver at the specified distance for different environments. Radiating outward from the sender, the reception rate variance is low for all receivers up to some distance. Beyond a certain distance, the variance increases suddenly, and successive nodes see distinctly different packet reception variance. This again is a very graphic illustration of the gray area phenomenon. It indicates not only that some nodes in the gray area can see pathological loss, but that those nodes also see time varying packet loss. This has an important implication for schemes that perform topology control by excluding low quality wireless links. It is important for such schemes to continuously measure link quality, since reception rates can vary significantly over larger time-scales.

We made the packet delivery measurement data sets available to the community. No small data set of this sort can claim to be representative; given that environment has significant impact on packet delivery. However, we believe the data sets still serve as a starting point to understand packet delivery in dense wireless sensor networks. The data sets have been used by other groups in CENS in simulation based evaluation of new protocols.

FUTURE DIRECTIONS

We will continue to implement a tool suite for current sensor network research and experimentation. “Digest” serves as an early warning system in our monitoring architecture by computing aggregate network properties on the background. The design goal is to continuously provide critical metrics of system health with good energy-efficiency and robustness characteristics. We plan to migrate to the new version of TinyOS, and develop support for a wider variety of metrics.

By using the data sets from our empirical study of radio characteristics, we also plan to develop new radio models and simulation tools to support more realistic simulation of densely deployed wireless sensor networks. For example, we intend to design and implement topology scenario generators to reflect the spatial and temporal characteristics of radio communication. Ricardo Hu (partially supported by NSF Undergraduate Research Experience Program) will start to develop and implement such tools within the EmStar framework.

Given the improved availability of real systems and testbeds such as James Reserve, we will further explore what types of data are important for the management of sensor networks. With the feedback from the real world,  we also intend to further explore how to efficiently represent and collect such management data.

PEOPLE

FACULTY

Prof. Ramesh Govindan

GRADUATE STUDENTS

Richard Hu
Jerry Zhao