balaji prabhakar departments of ee and cs stanford university joint work with: deepak merugu, naini...
Post on 15-Jan-2016
220 views
TRANSCRIPT
High PerformanceSwitching and RoutingTelecom Center Workshop: Sept 4, 1997.
Balaji PrabhakarDepartments of EE and CS
Stanford University
Joint work with:
Deepak Merugu, Naini Gomes, Gajanana KrishnaStanford University
N.S. Rama and many, many other Infoscions
It pays to do the right thing:Incentive mechanisms for decongesting the road
• Fuel and time costs in 2005, US-wide
A. $ 8 million
B. $ 80 million
C. $ 8 billion
D. $ 80 billion
Cost of congestion
• US auto bailout: $ 25 billion• Stimulus package for Wall Street: $ 700 billion
• Fuel wasted in urban U.S. in 2005
A. 3 million gallons
B. 30 million gallons
C. 3 billion gallons
D. 30 billion gallons
Fuel cost
• Equal to fuel consumed in all of the U.S. in 6 days !– 2006 data
• Congestion occurs when many agents simultaneously use a scarce resource; it is a standard feature of networked resources– Transportation Networks, Electric Networks, Computer Networks, Telephone
Networks, …
• Various methods have been developed for coping with congestion, usually tailored to the situation– Carpools, traffic-sensitive signal lights, tolls– Differential pricing for peak and off-peak use– Congestion control algorithms like TCP in the Internet– Congestion control standard for Data Center Ethernet: IEEE 802.1
• These methods usually address congestion, once it has occurred
• Incentive mechanisms, the topic of this talk, can postpone or even prevent the occurrence of congestion
Congestion management:Background
• Current methods: Charge drivers who enter “congested zone” – E.g. London, Singapore, Stockholm
• Our proposal: Charge congestors, pay decongestors – Deploy incrementally, no need for every one to start on day one
Congestion pricing
• “The right to congest" can be made into a tradable commodity– Those who use the roads at congested times pay those who stay away during such
times – This is incentive compatible: The congestor benefits from the reduced congestion;
the decongestor is appropriately compensated
Main Thesis
Load
Delay
0 1
• Small good deeds don’t carry adequate rewards, so they aren’t performed– A system, which pools individual rewards, but pays out a few large sums through
raffles may carry adequate incentives
• In games with low stakes, players are more risk seeking
Auxiliary Thesis
• The INSTANT (Infosys-Stanford Traffic) project is a pilot study of using an incentive mechanism to decongest road traffic– The goal is to incentivize Infoscions to commute at uncongested times to and from work– This leads to shorter commute times, reduced congestion, lower fuel costs and pollution
The INSTANT project
• The INSTANT project– Launched by N.R. Narayanamurthy, co-Founder and Chief Mentor of
Infosys on Oct 6, 2008 at Infosys, Bangalore– The program ran for 27 weeks and ended on April 10, 2009– Pays commuters money through a raffle mechanism for coming early
• 14,000 commuters eligible for the scheme• Results show that number of early comers has nearly doubled• About 1900 employees have been rewarded over the 6 months• The average prize money per winner is about Rs. 1400 ( or $28)• If the prize money was collected from everyone, the share per person is Rs. 7 (14 cents) per week.
The INSTANT project: Synopsis
INSTANT: Bus Data
• 120 starting points, 4 major routes• 240 buses• Commute data from January 2006 to June 2008
– Morning data presented (evening data available, similar patterns: leaving early saves a lot of commute time)
– Pick up, drop off times; bus occupancies
Bus #
Pick-up
TimeDrop Time Pick-up Point Cap A B C
54 700 732 JAYA NAGAR 4TH BLOCK /18TH MAIN 49 61 12
56 650 738 NANDIGARDEN / R.V.DENTAL COLLEGE 49 45 4
57 700 752 JAMBUSAVARIDINNE/R.V.DENTAL 49 38 11
60 700 745 GOTTEGERI / B.G.PARKING LOT 49 28 21
A - OccupancyB - StandingC - Empty Seats
Format of data
Monthly average numbers of commuters
Buses deployed to service demand
Bus occupancy: Early vs lateJune 2008; Bus capacity = 49
Over capacity
Journey times from Jayanagar
Adarsh Garden Minerva Circle
Other journey times
• LONG TIME, NO SEAT!
• There is a huge difference in commute times – A 7.15am bus into Infy is at least 30 mins quicker than an 8.15am bus– Similarly, a 5pm bus from Infy is at least 30 mins quicker than a 6.15 or 7.15pm bus
• Most Infoscions can benefit by coming in early and leaving early; the benefits are manifold:
– Shorter commutes by at least 1 hour– More comfortable rides (i.e. empty seats)– Savings in fuel costs (about Rs. 20,000 per day)
Summary of bus data analysis
• In addition to the above, we propose a monetary incentive mechanism to help transition to earlier commute times!
Commuter’s working day
begins
Commuter’s working day
begins
Arrival (swipe-in)
time
Arrival (swipe-in)
time
Arrival after 8.30 AM Arrival before 8.00 AM
No creditNo creditEarn 1.5 creditsEarn 1.5 credits
Incentive mechanismIncentive mechanism
Rewards givenweekly
Rewards givenweekly
The incentive mechanism At a Glance
Earn 1 creditEarn 1 credit
Arrival between8.00 – 8.30 AM
No. of rewards x Amount of reward (credits needed)
The reward pyramid•Total sum of money = 96,000•Total number of rewards = 66
2 x 12,000 (20)
4 x 6,000 (12)
12 x 2,000 (7)
48 x 500 (3)
An illustration
14 x 2,000 (7)
52 x 500 (3)
Results
Commuter arrivals
Pilot launched Buses rescheduled Pilot ended
Percentage of commuter arrivals
Pilot launched Buses rescheduled Pilot ended
Average morning bus commute time(and total person-hrs saved)
2000 p-hrs
1400 p-hrs
2300 p-hrs
100 person-hrs
500 p-hrs600 p-hrs
2600 p-hrs
2400 p-hrs2200 p-hrs
2000 p-hrs
Pilot launched Buses rescheduled Pilot ended
Adjustments to the bus schedule
• The INSTANT project advances a strong argument in favor of rewarding decongestors (and charging congestors)
– In addition to time and fuel savings, it led to the reduction of the Infosys fleet by 8 buses
• Next steps– Parking and Transportation Services (P&TS), Stanford University
Summary and next steps
• Parking at Stanford: Reduce peak hour trips
At Stanford