AbstractâThis paper contains a taxonomy of techniques for monitoring the energy consumption of electronic devices in communication networks. It furthermore ...
The 9th Annual IEEE Consumer Communications and Networking Conference - Demos
Energy Consumption Monitoring Techniques in Communication Networks Dominique Dudkowski, Konstantinos Samdanis NEC Laboratories Europe, Heidelberg, Germany Email: {dudkowski|samdanis}@ neclab.eu not require physically tampering with the measured device but make use of existing instrumentation support. A typical example is the retrieval of information via SNMP, which requires proper support in the management information base (MIB) of the measured device.
I. INTRODUCTION
C
Application Tier
electronic devices in communication networks with the objective of operating networks more energy-efficiently requires accurate knowledge about the energy consumption of the devices in the first place. We present a prototype that shows how the energy consumption of various types of electronic devices in a communication network can be obtained using different monitoring techniques. Specifically, our demonstration addresses heterogeneous environments with different types of electronic devices, including network equipment (e.g. switches, wireless access points, PBXs) and consumer devices (e.g. IP and legacy phones, laptops, PCs). In such environments a single monitoring technique is generally infeasible. We explore the space of possible monitoring techniques and describe important implications productive monitoring solutions must meet. ONTROLLING
Application n Device Control
Monitoring Interface Monitoring Tier
real-time energy consumption monitoring
persistent energy consumption data storage
Measurement Interface Measurement Tier
device management (e.g. plug-and-play integration)
instrumentation and measurement technique management
Fig. 1. High-level energy consumption monitoring platform.
Furthermore, two types of measurement techniques specify how energy parameters are obtained. Direct measurement of energy parameters refers to techniques that obtain energy metrics in an immediate way, such as the effective power in Watt or the aggregated energy consumption in kWh. Modelbased measurement of energy parameters denotes techniques that measure a device’s performance parameters and in a second step translate these into energy parameters by means of a device-specific energy consumption model. Both measurement and instrumentation techniques can also be combined (Fig. 2).
II. PROTOTYPE CONCEPT Conceptually, our prototype implements the platform in Fig. 1. Energy management applications, e.g. energy consumption visualization and power control functions, are located in the application tier. The measurement tier contains functions for integrating electronic devices in the measurement process and for obtaining a device’s energy or performance parameters. The monitoring tier implements real-time monitoring of energy parameters (e.g. for visualization) and their persistent storage (e.g. for offline consumption analysis). Monitoring the energy consumption is achieved by using various instrumentation techniques. Hardware-based instrumentation is implemented by introducing dedicated physical devices that are able to access energy parameters, for instance, a power meter in between the power outlet and the power connector of the measured device. Software-based instrumentation is implemented by using dedicated software agents that are deployed on the measured device, for instance, an agent that accesses the native API of a PC’s operating system. Noninvasive instrumentation refers to methods that do
instrumentation technique
modelbased
direct
network switch (in the role of a measuring device) monitoring the packet rate of attached devices (e.g. IP phone)
PC‘s CPU load obtained via Windows Performance API, translated locally into PC‘s effective power
switch‘s CPU load obtained via SNMP queries (MIB support required), remotely translated into switch‘s effective power
power meter, monitoring the effective power of attached network switch – PoE-enabled switch monitoring the effective power of attached devices
PC‘s effective power obtained via operating system‘s API
PoE-enabled switch‘s effective power obtained via SNMP queries (MIB support required)
hardware-based
software-based
noninvasive
Fig. 2. Matrix of instrumentation and measurement techniques.
54
measurement technique
978-1-4577-2071-0/12/$26.00 ©2012 IEEE
Application 1 Visualisation
...
Abstract—This paper contains a taxonomy of techniques for monitoring the energy consumption of electronic devices in communication networks. It furthermore presents a prototype that implements a combination of several techniques and derive a set of key implications for a productive monitoring platform.
observation that it is difficult to apply model-based measurement techniques in a multi-site environment. Fig. 4 shows a model mismatch for an energy consumption model that was calibrated for 230 V and then applied in a 110 V power grid. While the model’s accuracy was maintained for low and high CPU utilization, a significant deviation from direct measurement could be observed for CPU loads around 50%. This fact emphasizes the sensitivity of model-based energy consumption monitoring with respect to power supply characteristics and complicates the use of energy consumption models in multi-site environments for which additional calibration is required to obtain accurate measurements.
III. PROTOTYPE IMPLEMENTATION We have implemented a prototype of an energy consumption monitoring platform (Fig. 3) and tested it in a 230 V (EU) as well as a 100 V (Japan) power grid.
real-time energy consumption monitoring application and database for persistent storage
power meter: hardwarebased direct monitoring
Power-over-Ethernet switch: non-invasive direct monitoring
Ethernet
Windows PC: software- and modelbased monitoring
Power-over-Ethernet devices: hardware-based direct monitoring
model mismatch
Fig. 3. Prototype of the energy consumption monitoring platform.
The main component of the setup is a switch enabled with Power-over-Ethernet (PoE) to which a number of PoE-capable devices are connected. Two PCs act as measurement subject and host for a monitoring application (Fig. 4). Hardware-based direct monitoring (Fig. 2) obtains the PC energy consumption to compare it with a software-based/model-based monitoring approach on the monitoring PC that was realized with a software agent making use of operating system-specific performance libraries. An energy consumption model translates the PC CPU load into the PC energy consumption. An SNMP-based noninvasive monitoring technique was used on the network switch to retrieve the switch’s instant effective power. Direct hardware-based monitoring was used for the devices attached to the PoE switch (IP phone, wireless access point), where the switch acts as the measuring device. Fig. 4 shows a screenshot of the graphical user interface of our real-time monitoring application, situated in the application layer of the monitoring platform (Fig. 1). The interface shows the power meter’s reported effective power of the PC, and the estimated power based on CPU load. Our prototype demonstrates various characteristics of energy consumption monitoring in 230/110 V power grids from which we obtain the following key results: Manageability and energy penalty: Instrumentation by hardware-based (power meter) and software-based techniques (software agents) has shown difficult to manage. Software agents are OS-dependent and require specific knowledge about how to retrieve suitable input parameters for modelbased monitoring techniques. Both instrumentation techniques come at the cost of additional energy consumption overhead: the power meter consumes roughly 2 Watts while connected, which is equivalent to the switch’s power when idle. Modelbased measuring techniques require additional CPU utilization, which leads to additional energy consumption. Energy consumption model accuracy: The deployment of the platform in different types of power grids has led to the
Fig. 4. Energy consumption monitoring application.
Accessibility, security, and privacy: While physical access to productive ICT environments is generally restricted, our prototype confirms that only direct noninvasive monitoring is generally feasible. Specifically, direct instrumentation, e.g. by means of the power meter, raises significant security questions as it requires to interfere with a running system’s power grid (e.g. the physical hosts of a Web service provider). While the introduction of measurement equipment may be possible when uninterruptible power supplies (UPS) are in place, introducing monitoring techniques that require hardware or software to be deployed is not appropriate in dynamic environments for the sole purpose of energy consumption monitoring. With respect to privacy, the prototype showed that from a single parameter (the CPU load in the considered case) it is possible to deduce a user’s PC activity. Additional correlations between a complete network’s configuration and the contained devices’ energy consumption allows to derive even more detailed user activity patterns. Privacy constraints therefore require explicit security solutions in a viable energy consumption monitoring system. IV. CONCLUSION This paper elaborated a prototype for energy assessment, showing how the energy consumption of various types of electronic devices can be obtained by different instrumentation and measurement techniques. Future work will address the implementation of management functions that are able to control individual electronic devices in terms of different power states (e.g. ON, STANDBY, OFF).
55
