Transcript ppt

A Congestion Pricing User
Study Using a 802.11a
Wireless LAN
Jimmy Shih,
Randy Katz,
Anthony Joseph
Problem Statement
 How to allocate wireless bandwidth?
 Insufficient bandwidth during peaks.
 Use prices to encourage some users to conserve
bandwidth during congestion.
• E.g. Encourage heavy users to conserve bandwidth
to support more light users.
 Congestion pricing:vary prices according to load.
• Allocate scarce bandwidth.
• Save over-provisioning.
• Provide guaranteed service.
 Goals:
 Measure effectiveness of using congestion pricing.
 Understand user acceptance to changing prices.
State-of-the-Art
 Proposals of different congestion pricing
mechanisms.
 Senders pay at each congested point.
 Senders pay to neighboring domains.
 Receivers pay for received congested packets.
 Simulations:
 Assume particular user model.
 Show that congestion pricing potentially can be better.
 Little actual experimentation.
 Need to quantify tradeoffs of congestion pricing.
 Need to demonstrate user acceptance.
Methodology
 Combination of user
experiments and
simulations to quantify
effectiveness of
congestion pricing for
wireless bandwidth.
1. User Experiments for Modeling Users’ Reactions to Price Changes
Price
Small Number of Reactions to
Price Changes
Real Users
2. Simulations for Managing Under High Loads
User Model
Behavior Workload
Model
Model
1. Experiments for understanding
Load
Large Number of
Price
Simulated Users
acceptance and reactions to
Usage
price changes.
2. Simulations for understanding 3. User Experiments & Simulations for Verifying Users’ Reactions
management of congestion
Usage
pricing under high loads.
Large Number of
User Model
Simulated Users
3. Combining user experiments
Rules for Managing
Load Price
with simulations to verify
Under High Loads
Price
users’ reactions and tradeoffs.
Small Number of Reactions to
Real Users
Usage
Price Changes
Prototype
Web
Server
1
Internet 2
Internet3
Access
Router
1:Control: TCP connection.
2:Control: web connections.
3:Data
WLAN
Packeteer
PacketShaper
User Interface
 Rate-limit each IP address
to certain bandwidth.
 Give each user some free
tokens a week.
 User needs more
bandwidth, goes to web
server, looks at prices, and
decides to purchase more.
 Provide users with
current usage
information.
 Users can change their
bandwidth selections at
any time.
Pilot Study
 Tested out UI and
stability of the
prototype with
graduate students
using a wired
Ethernet network.
 Traces of download
and upload
bandwidth usages.
 Download
 Upload
Spring Semester Plan
 Deploy a 802.11a network using 10
access points around Soda Hall.
 Provide 50 graduate students with
wireless cards for participating in the
pricing experiments.
 Expected Results:
 Show users willing to interact with UI that
only requires their attention when they need
more bandwidth.
 Measure percentage of new and existing users
requesting excess bandwidth when prices
change.
Summary
 Determine effectiveness of using dynamic
prices for allocating scarce bandwidth.
 Conduct a user study using a 802.11a WLAN.
 Expected Contributions:
 Collect traces of wireless bandwidth usages.
 Understand user acceptance to changing
prices.
 Model users’ reactions to changing prices.
 Quantify effectiveness of congestion pricing
for wireless bandwidth.