Title: Power-proxying on the NIC : a case study with the Gnutella file-sharing protocol
CITATION PDF VIEWER THUMBNAILS PAGE IMAGE ZOOMABLE
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/UF00094698/00001
 Material Information
Title: Power-proxying on the NIC : a case study with the Gnutella file-sharing protocol
Physical Description: Book
Language: English
Creator: Purushothaman, Pradeep
Navada, Mukund
Subramaniyan, Rajagopal
Reardon, Casey
George, Alan D.
Publisher: Purushothaman et al.
Place of Publication: Gainesville, Fla.
Copyright Date: 2006
 Record Information
Bibliographic ID: UF00094698
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.

Downloads

This item has the following downloads:

LCN2006b ( PDF )


Full Text









Power-Proxying on the NIC:
A Case Study with the Gnutella File-Sharing Protocol*



Pradeep Purushothaman, Mukund Navada, Rajagopal Subramaniyan,
Casey Reardon, Alan D. George
High-Performance Computing & Simulation (HCS) Research Laboratory
Department of Electrical and Computer Engineering, University of Florida
Gainesville, Florida 32611-6200
{purushothaman, navada, subraman, reardon, george})@hcs.ufl.edu


Abstract

Edge devices such as desktop and laptop computers
constitute a majority of the devices connected to the
Internet today. Peer-to-Peer (P2P) file-sharing
applications generally require edge devices to
maintain network presence whenever possible to
enhance the robustness of the file-sharing network,
which in turn can lead to considerable wastage of
energy. We show that energy can be saved by
p,, m,1,oi edge devices to enter into standby state and
still maintain network connectivity by proxying
protocols in the Network Interface Card (NIC).


1. Introduction

A Peer-to-Peer (P2P) network is a distributed
system that breaks the traditional client-server
computing paradigm as nodes can act as both a client
as well as a server. One of the popular applications of
P2P networks is file-sharing. Gnutella, Bit Torrent,
FastTrack, SoulSeek, etc. are a few examples of
popular P2P file-sharing protocols. The topology of a
P2P file-sharing network is largely dynamic with
nodes joining or leaving the network randomly. P2P
file-sharing is attractive with client/server-based, file-
sharing due to the fault tolerance property of P2P
networks. Since multiple nodes on a P2P file-sharing
network usually maintain copies of a single file, file-
sharing is not necessarily interrupted as a result of
nodes leaving the network. However, P2P file-sharing
protocols require nodes to maintain network presence
whenever possible to provide this robustness.


This material is based on work supported by the National Science
Foundation under Grant No. 0519951.


Power consumption of edge devices connected to
the Internet will be major cause of concern in the near
future [1]. It has been estimated that a desktop is
generally left idle for 75% of the time after it is
powered on. To conserve power, operating systems
(OS) usually incorporate a low-power standby state.
Edge devices enter this state after a period of idle
activity defined by the user. Transitioning into the
standby state usually disables the Network Interface
Card (NIC) resulting in disruption of network
connectivity. Such interruptions could affect network
applications such as P2P file-sharing. Consequently,
system users are forced to disable standby settings.
This paper proposes a new method called po\cil-
proxying" that refers to selectively proxying a subset
of protocol semantics of a network protocol through a
separate hardware controller on the NIC. Proxying
helps increase the amount of time an idle edge device
spends in the standby state without compromising
network connectivity. This NIC with the hardware
controller for proxying is called a Silllt" NIC as it
intelligently wakes up the host when it receives a
packet that cannot be proxied. The idea is validated by
considering Gnutella as a case study and obtaining an
estimate of power savings.

2. The Gnutella Protocol

The current version of this protocol, Gnutella V0.6,
consists of the following descriptors: PING, PONG,
QUERY, QUERYHIT, and PUSH. The PING
descriptor is used to actively discover hosts on the
network. A servant (Gnutella node) receiving a PING
descriptor is expected to respond with one or more











PONG descriptors. QUERY is the primary mechanism
for searching for files in the P2P network. A servant
receiving a QUERY descriptor will respond with a
QUERYHIT if it shares the requested file.

3. Experimental Setup

The experimental setup to determine the energy
savings obtained by proxying Gnutella consists of two
steps: trace collection followed by simulation of the
operation of a power-proxy in a Gnutella network.
Trace collection involves collecting statistics on the
received Gnutella packets at an edge device. The
collected trace is then used as input to a simulator,
designed and built to study and estimate the amount of
power savings that could be obtained. The Gnutella
network bandwidth as seen from the edge device can
also be varied to study the influence of node bandwidth
and standby time on obtained power savings.
The simulator consists of a peer-network entity that
abstracts a Gnutella network and an edge device with a
SiniIl" NIC connected to the peer network. The peer
network issues Gnutella packets to the edge device
based upon the input trace. When the edge device is in
standby state, all the Gnutella packets except the file
upload requests are handled by the proxy.

4. Results and Inference

Traces show that average inter-arrival time for 90%
of PING, PONG, and QUERY packets is within 90
sec, 4 sec, and 30 sec, respectively. These
observations show that Gnutella network is not really
idle even on a passive node (a node that does not
initiate queries).
Simulation of the P2P network was performed using
the simulator described in Section 3. The standby time
for the host was varied from 150 sec to 600 sec in steps
of 150 sec, and the bandwidth of the edge device was
varied from 100 Kbps to 100,000 Kbps. Figure 1
shows that an edge device in the Gnutella network is in
the standby state for more than 85% of time as a result
of proxying, irrespective of network bandwidth. For
high-bandwidth nodes, increasing standby time settings
only reduces the amount of time spent in standby state,
as the node waits a longer time to enter standby state.
Figure 2 shows the inter-arrival time for file-upload
requests for an edge device in the Gnutella network.
As shown in the figure, for file upload requests at a
peer with 100 Mbps bandwidth and standby time of
300s, 60% of the inter-arrival times are one hour or
longer. Thus, proxying Gnutella can result in the edge
device spending more time in the standby state, as only
file-upload requests need to be handled by the host.


98



92

90
-





88 150 sec -
300 sec - -
450 sec --* ,.
800 sec ---
86
100 1000 10000 100
Bandwidth (Kbos0
Figure 1: Analysis of Gnutella Proxying


2000 6000 10000 14000
Inter-Arrival Time (sec)
Figure 2: File Upload Request


5. Conclusions

Proxying is a suitable technique to reduce power
consumption in networked edge devices. Analysis of
the collected traces justifies the need for proxying and
simulation results show that a substantial amount of
power can be saved by proxying Gnutella.
Implementing the power-proxy on a NIC can definitely
achieve significant energy savings as it allows the rest
of the system to remain in low-power standby mode.
Future work would focus on various architectural
design strategies and tradeoffs for realizing the proxy
in hardware. A prototype NIC with proxy
functionality will also be developed to quantify the
power savings.

6. Reference

[1] K. Christensen, P. Gunaratne, B. Nordman, and A.George
"The next frontier for communications networks: power
management," Computer Communications, Vol. 27, No. 18,
pp. 1758-1770, December 2004.




University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs