Dr Scott Watterson, Dr David Lowenthal, Dr Kang Li , Haijin Yan, Rupa Krishnan
With the explosion of battery-constrained mobile devices, conserving energy has become increasingly important. One significant source of consumed energy on such devices is the wireless network interface card (WNIC). This is especially true in situations where data is received for a long period of time, e.g., file downloads and streaming multimedia applications. These applications require the WNIC to remain in idle, receive, or transmit mode, all of which use significant amounts of power. The project involves reducing the WNIC power consumption, by turning it to a lower power mode during periods of no data activity - during a http or ftp download.
This problem can be approached in different ways. One way would be to have a proxy buffer the data coming to the client [1]. This is similar to the IEEE 802.11 power saving mode. In this protocol, the client periodically turns the NIC down. The access points buffers the data for the client, when it is switched off. The client turns the NIC up every beacon period to retrieve the data. An improvement to this protocol is the bounded slowdown protocol [2]. As a part of this project we have emulated the 802.11 protocol as a transparent proxy to compare against other client side algorithms. A schematic diagram of the emulation of the 802.11 as a proxy is given below.
Fig 1. Emulation of 802.11 in a transparent proxy (Access Point is the proxy)
The goal of this work is to allow mobile clients to save energy during TCP downloads in a client centered manner, without any assistance from servers, proxies or IEEE 802.11b power saving mode. We focus our efforts in this work on handling TCP streams initiated by the client to a server via request/response, using round-trip time estimates to determine when to transition the WNIC between idle and sleep mode.
One common use of mobile device is browsing the internet. The goal is to save power during a web download. This is done by maintaining the state of each client initiated connection. When all connections are idle the client transitions the NIC into low power mode. Results show that this scheme saves on an average 25% to 29% energy with very little increase in transmission time.
For large files (e.g. Gnutella downloads), saving energy in the WNIC of a client is best when bursts of data are transmitted in such a way that the client can transition its WNIC to sleep state between bursts. However, TCP does nothing to combine packets into bursts—instead, it attempts to smooth the packet stream—which makes saving energy during long transfers difficult. We have designed and implemented a technique for TCP downloads in mobile clients that trades reduced WNIC energy for increased transmission time for large files. First, the client shapes the traffic of files sent by the server to increase gaps between packet arrival times, increasing time that the WNIC can be placed in sleep mode. This is done by modifying the client side advertised window for all except the last acknowledgement of the window. Second, a client estimates arrival times for packets using prediction of round-trip times by leveraging TCP timestamp information. Finally, a client predicts the end of the window using estimates of packet interarrival times and deviations. The traffic shaping is performed strictly by the client, using TCP information to manipulate how the data is sent by the server.
As a part of the project we have also studied the effect of the bursting client on routers and other clients of the network. The simulation was done on ns-2, where the number of bursting clients in the network was varied and the effect on the bottleneck routes was studied. (The ns-2 code for bursting client will be put up for download shortly).
1. " Client-Centered, Energy-Efficient Wireless Communication on IEEE 802.11b Network"
Haijin Yan, David Lowenthal, Kang Li, Rupa Krishnan, and Larry Peterson. To appear in IEEE Transaction on Mobile Computing, Spring 2006.
2. " Client-Centered Energy*Delay Reduction for TCP Downloads"
Haijin Yan, Rupa Krishnan, Scott Watterson, David Lowenthal, Kang Li, and Larry Peterson. To appear in IWQoS 2004.
3. " Client-Centered Energy Savings for Concurrent HTTP Connections "
Haijin Yan, Rupa Krishnan, Scott A. Watterson, David K. Lowenthal. To appear in NOSSDAV 2004.
[1]
Surendar Chandra. Wireless network interface energy consumption implications of popular streaming formats. In MMCN, Jan 2002.[2] R. Krashinsky and H. Balakrishnan, "Minimizing energy for wireless web access with bounded slowdown," in
MobiCom 2002, Atlanta, GA, September 2002.