Some pricing models for the Diff Serv architecture

  • Slides: 19
Download presentation
Some pricing models for the Diff. Serv architecture Y. Hayel 18/09/2020 workshop prixnet

Some pricing models for the Diff. Serv architecture Y. Hayel 18/09/2020 workshop prixnet

Outline n n n Introduction to Diff. Serv architecture Two priority pricing schemes An

Outline n n n Introduction to Diff. Serv architecture Two priority pricing schemes An edge-pricing : the Cumulus pricing scheme Congestion-based pricing with priority service Conclusions 18/09/2020 workshop prixnet

The Diff. Serv architecture (1) Why ? quality of service differentiation we want to

The Diff. Serv architecture (1) Why ? quality of service differentiation we want to have « Better than best effort » for the management of traffic n How ? different priority classes (PHBs) are assigned at the edge of the backbone for an aggregation of traffic packet routing policy in the backbone depend on the class n 18/09/2020 workshop prixnet

The Diff. Serv architecture (2) backbone Edge router 18/09/2020 workshop prixnet

The Diff. Serv architecture (2) backbone Edge router 18/09/2020 workshop prixnet

The Cumulus Pricing Scheme (CPS) n n Proposed by Reichl et al. (2001) Simple

The Cumulus Pricing Scheme (CPS) n n Proposed by Reichl et al. (2001) Simple model based on a contract dealing with Qo. S parameters like bandwidth between the provider and users. Reaction rule in terms of cumulus points renegociation of contract penalty to provide correct economic incentives n 18/09/2020 workshop prixnet

Management Points are assigned to user according to the following rule: or n with

Management Points are assigned to user according to the following rule: or n with The ISP estimates user’s consumption from measurements cost of measurements for a given precision n 18/09/2020 workshop prixnet

+2 +5 0 0 +4 +1 7 7 7 11 1 -1 -3 +1

+2 +5 0 0 +4 +1 7 7 7 11 1 -1 -3 +1 +2 0 0 0 V(t) x 1 x 0 Total cumulé 18/09/2020 2 workshop prixnet 0 -3 -2 t

Choosing the thresholds User must respect his contract to provide network management optimization of

Choosing the thresholds User must respect his contract to provide network management optimization of users level of satisfaction constraint on negative thresholds independent of the Utility function n Optimization of the network revenue thresholds values under constraint that each user respects his contract n 18/09/2020 workshop prixnet

In Diff. Serv architecture n n n CPS is tailored to complement Diff. Serv

In Diff. Serv architecture n n n CPS is tailored to complement Diff. Serv in the provisioning of Qo. S Contract is done at the edge of a Diff. Serv domain CPS contract could be specified in the SLA 18/09/2020 workshop prixnet

Pricing multiclass networks Different priority approaches: - Posted price and simulation - Posted price

Pricing multiclass networks Different priority approaches: - Posted price and simulation - Posted price and non-cooperative game - Non-posted price and queueing models within cost incurred by users is decomposed into price and delay 18/09/2020 workshop prixnet

Posted pricing (Cocchi et al. ) Two levels of priority for each packets :

Posted pricing (Cocchi et al. ) Two levels of priority for each packets : - priority service flag - priority no-drop flag (if there is no room in the queue) n n n The authors compare a flat per-byte price and a priority pricing based on the priority flags. Given valuation functions for each application and set the prices are determined by simulation optimizing an utility function (valuation minus cost) 18/09/2020 workshop prixnet

Another model (Marbach) Prices per sent-packet are static. n Time discretized in slots -

Another model (Marbach) Prices per sent-packet are static. n Time discretized in slots - higher priority classes served first - packets not served in a slot are lost n Analysis by a non-cooperative game, where each user chooses the allocation among classes which maximize his net benefit. n Proof of existence and uniqueness of an equilibrium. n Somes extensions to dynamic game and bursty traffic are performed by the author in other papers. n 18/09/2020 workshop prixnet

Model n User r throughtput is n User r chooses allocation such that n

Model n User r throughtput is n User r chooses allocation such that n The Utility function is suppoped to be increasing, bounded, stricly concave and twice differentiable. 18/09/2020 workshop prixnet

Non-posted pricing n n n We have n classes and class-i jobs arrive with

Non-posted pricing n n n We have n classes and class-i jobs arrive with rate their service rate is V( ) is the total expected gross value gained by the system so that: unit-price charged to a class-i job that we want to optimize with arrival rates 18/09/2020 workshop prixnet

Model n n The class-i demand relationship is: is the cost by unit of

Model n n The class-i demand relationship is: is the cost by unit of delay and is the expected waiting queueing delay of a class-i job (known for a M/M/1) 18/09/2020 workshop prixnet

Goals n Maximization of the expected net value: optimal arrival rates with the first

Goals n Maximization of the expected net value: optimal arrival rates with the first order condition of the previous equation optimal vector price with the demand relationship 18/09/2020 workshop prixnet

A. S. version n n We apply the Aumann-Shapley cost sharing mechanism to the

A. S. version n n We apply the Aumann-Shapley cost sharing mechanism to the mean number of jobs in the system. We have the budgetbalanced property (sum of the cost for all classes is equal to the total cost). We assume that the class-i demand relationship is: The unit valuation for each class is decomposed into price and uniform mean of delay incurred by its jobs. We also obtain optimal rates and prices. 18/09/2020 workshop prixnet

Conclusions and perspectives n n 1. 2. Diff. Serv could be the future Internet

Conclusions and perspectives n n 1. 2. Diff. Serv could be the future Internet architecture to provide Qo. S Our aims: Designing new models; improving existing ones Compare analytically and experimentally (on the testbed) a maximum of models 18/09/2020 workshop prixnet

References n n P. Reichl et al. « How to overcome the feasibility problem

References n n P. Reichl et al. « How to overcome the feasibility problem for tariffing internet services: the Cumulus Pricing Scheme » , in proc. of IEEE ICC 2001. R. Cocchi « Pricing in Computer Networks: Motivation, Formulation, and Example » , IEEE/ACM Transactions on Networking, 1993. P. Marbach « Pricing Differentiated Services Networks: Bursty Traffic » , IEEE Infocom 2001. H. Mendelson « Optimal Incentive-Compatible Priority Pricing for M/M/1 Queue » , Operations Research, 1990. 18/09/2020 workshop prixnet