Two-Market Inter- domain Bandwidth Contracting Anusha Uppaluri, University of Nevada, Reno ([email protected]) Praveen Kumar, Rensselaer Polytechnic Institute Murat Yuksel, University of Nevada, Reno Aparna Gupta, Rensselaer Polytechnic Institute Koushik Kar, Rensselaer Polytechnic Institute Industrial Engineering Research Conference, IERC, 61st Annual Conference & Expo, Reno, Nevada, May 21-25, 2011. 1
Two-Market Inter-domain Bandwidth Contracting. Anusha Uppaluri, University of Nevada, Reno ([email protected]) Praveen Kumar, Rensselaer Polytechnic Institute Murat Yuksel, University of Nevada, Reno Aparna Gupta, Rensselaer Polytechnic Institute - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Praveen Kumar, Rensselaer Polytechnic Institute Murat Yuksel, University of Nevada, Reno
Aparna Gupta, Rensselaer Polytechnic InstituteKoushik Kar, Rensselaer Polytechnic Institute
Industrial Engineering Research Conference, IERC, 61st Annual Conference & Expo, Reno, Nevada, May 21-25, 2011.
1
Overview• Introduction• Problem Formulation
– Network Model– Single Edge – to – Edge Contract Link– Multiple Edge – to – Edge Contract Links
• Simulation Experiments– Single Edge – to – Edge Link– Multiple Edge – to – Edge Links
• Network Setup• Results
• Conclusion and Future Work
2
Introduction
• Migration of data rich applications indicate that the increasing demand for bandwidth is unlikely to vanish
• Managing bandwidth allocation to customers presents a major challenge and received a lot of attention
• Many approaches involving advanced techniques for achieving efficient contracting among ISPs have been explored
3
Introduction (contd..)
• Crucial handicap of existing inter – ISP economics is coarse granularity of contracts
• Precursor to realization of more dynamic & automated contracting is sufficient motivation for ISPs to invest and install necessary tools & protocols
• Question focused on : how much benefit there is if the contracts were classified into simple two market regime, highly dynamic and long time scale durable
4
Introduction (contd..)
• Also, an important issue is the management of risks in ISPs investments
• Though an automated and dynamic way of establishing contracts will enable ISPs to flexibly allocate their resources several questions arise:– When and where to advertise a new contract– How to assess the risks involved and reflect
them on contracting parameters – How to divide links between different types
of contracts
5
Introduction (contd..)
• Model and formulate revenue maximization problem by considering the constraints imposed by correlation among contract links
• Through a detailed simulation several insights on optimal reservation levels for the two types of contracts are provided
6
Problem formulation
Contract switched Internet architecture considered7
Problem formulation(contd..)
• Edge – to – edge contract links are advertisable contracts between pairs of ingress and egress routers
• The Capacity of a contract link is equal to the minimum of the capacities of the physical links used to construct the contract link
8
Problem formulation(contd..)
• Capacity of the contract links must be segmented between short – term and long – term contracts
• There is enough demand for long – term contracts while the demand for short – term contracts is stochastic
• Revenue from unit demand for short – term contracts per unit time is higher than Revenue from unit demand for long – term contracts per unit time
9
Single Edge- to- Edge Contract Link
• Single edge – to – edge contract link is considered• Enough demand for long – term contracts is
assumed. Bandwidth allotted for long – term contracts is always used
• Bandwidth available for short – term contracts may or may not be used
• ISP must choose optimal value for bandwidth reserved towards long – term contracts so that total expected revenue per unit time is maximized
10
•
Single Edge- to- Edge Contract Link(contd..)
11
Revenue obtained from long - term
contracts per unit time
Revenue obtained from short – term contracts per unit
time
Bandwidth reserved for long – term
contracts
Capacity of the contract
link
Multiple Edge- to- Edge Links
• Assuming several edge – to – edge links in ISP’s network
• Edge – to – edge contract links share physical links which impose constraints on bandwidth reserved for long – term contracts of these contract links
12
•
Multiple Edge- to- Edge Links (contd..)
Bandwidth reserved for long –
term contracts
Capacity of the physical links used to construct the
contract link
13
•
= Revenue form long term contracts per unit time
Multiple Edge- to- Edge Links (contd..)
Revenue from long –term
contracts per unit time
Sum of revenue obtained from long term
contracts on multiple contract links
14
•
Multiple Edge- to- Edge Links (contd..)Revenue
obtained from short term
contracts per unit timeMinimum of
Short term demand and
residual capacity
Reservation for short – term contracts on contract linkReservation for short – term
contracts on a contract link
Remaining capacity of
physical link l
15
Multiple Edge- to- Edge Links (contd..)
•
16
Revenue from long – term
contracts per unit time
Revenue from short – term contracts per
unit timeBandwidth reserved
towards long term contracts
Capacity of contract link
Bandwidth reserved for long –
term contracts
Capacity of the physical links used to construct the
contract link
Simulation Experiments
Single edge – to – edge link• Capacity of single edge – to – edge contract link is
set to 10• Two distributions of short term demand
– Uniform distribution between 0 and 10– Truncated Gaussian distribution in the interval
[0,10] with mean 5 and standard deviation 1
17
Simulation Experiments(contd..)
18
Optimal long term reservation on contract link is shown. PL is increased from 0 and PS is fixed at 10
Optimal long term reservation on
contract link on Y-axis
PL is being increased from 0 to
10 on X-axis
Graph for truncated Gaussian
distribution
Graph for uniform
distribution
Multiple Edge – to – Edge Links
Network Setup• Real topology map of GEANT ISP with 23 routers
is used. 3 ingress & 3 egress routers are chosen. Nine edge- to- edge contract links are considered
• Capacity of physical links is set to 10 and so the capacity of contract link (Bie) is 10.
• Short term demand is taken to be uniformly distributed between 0 and 10
19
Results
Maximum total revenue, maximum long term revenue and maximum short term revenue are shown. PS is set to 10
20
PL is being increased
from 1 to 10 on X-axis
Maximum total revenue on Y-
axis
PL≤5 then PL is not high enough to
generate significant long term revenue
PL>5 then long – term contracts
generate higher revenue
Optimal reservation levels on nine links when PL=5 and PS=10 & PL=10 and PS=10
Results(contd..)
21
PL=5 and PS=10 PL=10 and PS=10
Contract links listed on X- axis
Contract link capacity on Y-axis
Contract links listed on X- axis
Contract link capacity on Y-axis
Increasing PL increases long -
term reservation levels
Long term reservation levels on 1,6,7 remain zero at all times
Conclusion and Future Work
• Revenue maximization problem was formulated for an ISP which wants to participate in two segments of bandwidth markets
• The level and distributional characteristics of short – term demand and interactions among contract links are key determinants
• In the future, possibility of multiple paths underlying a given single contract link instead of single path will be considered