PRICING MODEL FOR 3G/4G NETWORKS

Transcription

1 PRICING MODEL FOR G/G NETWORKS Eero Walleniu, Timo Hämäläinen Nokia Oyj IP Mobility Network (IMN) Hatanpäänvaltatie, P.O. ox 78, FIN- Tampere, Finland Phone +8, Univerity of Jyväkylä, Department of Mathematical Information Technology, P.O. ox, FIN- Jyväkylä, Finland, Abtract - Pricing of the future multimedia ervice in the G/G network will play a key role from operator point of view to achieve the maximum revenue and maximizing ROY. On the other hand pricing of the variou new ervice i a very important iue to ubcriber and epecially the pricing veru the acceptance of ervice will be a very delicate and important matter that mut be dealt very gently. Thi paper introduce model for the G/G ervice pricing including QoS. Keyword - G, G, QoS, Pricing, illing I. INTRODUCTION Thi work preent a pricing model for the future wirele network. It alo decribe a number of idea for G/G billing from the operator and ubcriber point of view. It define a model how operator can adjut their network load by uing a dynamic pricing chema and adjuting it to the local ubcriber acceptance curve by uing linearity factor L(x). Our imulation model can be ued for network pricing imulation and ervice charging tudie. The main part of thi tudy i the definition of the linearity factor and the uage of it for ervice pricing and cutomer acceptance adaptation. There are everal way to approach the pricing. To ue maximization technique on utilitie [McKie], [Jia], [Low], [Ji] where the utility i a meaure of atifaction while utilitie map a reource amount to that atifaction value. Alo multiple auction baed pricing mechanim have been preented a in [Che], [ic]. Compared to other tudie thi tudy bring a new view to the ervice pricing and taking care of the ubcriber acceptance and atifaction. It alo propoe method how to deal and olve the inequality between caual voice ervice uer and a heavy multimedia and volume uer. II. QOS AWARE PRICING In cae a ubcriber make a QoS change or the network require a QoS downgrade due to HO (Hand Over) or other reaon the price and QoS factor can be combined together by uing the eq.. P cci = Prij ± PQ pijk () Where: P cci = Total price for a call with known QoS Parameterization, P rij = aic rate for a default parameterized call, PQpijk = Change of the price due to a QoS property change, i = Subcriber cla (Gold, Silver, ronze and Economy) and j = GPP traffic cla. Table how QoS attribute and their value for each GPP traffic cla. Thi table alo include price factor/ change (PQ pij ) that can be expreed a in eq.. Table. Pricing of QoS attribute change QoS attribute Converational cla nominal/ change Streaming Interactivground ack- cla nominal/ cla cla change nominal nominal/ / change change III. LINEARITY FACTOR factor/ change* (PQ pij ) andwidth kb/ kb/ N/A N/A ±x/ * P * L(x) kb/ End- m m N/A N/A ±x/* To- P * L(x) End m delay ER ± x/ - 6(-7) * P * L(x) error x = number of QoS attribute tep, * Scaling, ee [TS.7]. PQ pij = ( + x / Qbij ) * Pij * L( x) () Where: x = amount of increae or decreae of a pecific QoS attribute value, Q bij = Nominal value of the QoS attribute in operator network, P ij = aic price of a call or connection in pecific traffic and ubcriber cla, i = GPP traffic cla, j= ubcriber cla indicator and L(x) = linearity factor. The linearity factor L(x) define if the pricing i linear function a een of the change of the QoS parameter. Divider ued in calculation correpond the maximum value of the ingle QoS attribute. The maximum value for the bandwidth i Mb/ and the correponding divider i kb/. The divider for the end-to-end delay i meaning //$7. IEEE PIMRC

2 maximum delay of m and for ER the divider i 6 (or 7) depending whether the traffic i RT or NRT. Uually the full linearity produce very high price at the top or very low price in the tart meaning that the voice call and multimedia call price ratio can be a high a : which hould be conidered to be too much. Linearity factor can be expreed a eq. : *x L( x) = A* ( e e ), x max.qos attribute value () Where: x, A and are linearity parameter and they hould be defined o that the price ratio between min and max QoS doe not raie unreaonable high. The Fig. illutrate behavior of the linearity factor. Linearity factor F(x) Linearity factor /A* (exp()-exp(-x)), A= and =.8 to.7 Fig.. Linearity factor behavior = =.7 =. =.8 The factor A et the bae level of the price and the factor adjut the lope of the curve, which determine the deduction of the price when uing high bit rate or volume tranfer. The relation to the cutomer claification can be expreed in the ame way. The aumption ha been that the quality tep can be equal in each traffic and cutomer cla. In the real life, however ome jutification may be needed. Value preented here are merely etimate for G/G network blocking and dropping propertie and they have not been verified yet in real operating environment. However, they preent a good bae for a prioritized network quality. IV. PRICE FACTOR The price factor i the variable, which i ued for optimizing the admiion control functionality. The goal i to optimize the income of the network ytem with the predefined threhold value for: o Maximum price for each ubcriber cla o Minimum bandwidth for each ubcriber cla in congetion ituation. andwidth price change tep (± PQ aij ) will be aumed to be ± % of the original minimum bandwidth and bae price. o Maximum delay for ubcriber cla. Delay price change tep (± PQ Dij ) will e aumed to be ± % of the original maximum delay and bae price. o Maximum ER for ubcriber cla. ER price change tep (± PQ eij ) will be aumed to be ± % of the original maximum ER and bae price. Pricing unit in the imulator will tune the admiion control according to thee value maximizing the total income. A. Network load dependent price change In different load ituation load dependent price change can be claified, dicrete or continuou. The Fig. how the continuou pricing principle. Continou pricing Linearity parameter L(x) Fig..Continuou pricing The continuou call pricing (bandwidth or volume baed) can be expreed a eq. : Pb = Aij * ( e e ) * Tl / () Where: P b = Unit price of the call or connection, A ij = linear price factor for each traffic and ubcriber cla, (e e -x* )= linearity factor and T l = Traffic load. Fig. depict ome reult of the claified pricing cheme Claified pricing Linearity parameter x Fig.. Claified pricing The claified call pricing can be expreed a equation a and b: P = /( A * ( e e ) * T /), A ij = C when < T l ij < T h % (a)... P = /( Aij * ( e e ) * Tl /), Aij=C N when < T l < % (b) and x = to 9 and = l

3 Where: P = Unit price of the call or connection, A ij = Contant price factor per ubcriber and traffic cla, (e e -x* ) = linearity factor, T l = Traffic load, C N = Contant price per ubcriber and traffic cla and T hl = Traffic load threhold value for price change. The linearity factor can alo be included into the C N, but for clarity it i hown here eparately. Ued pricing method hould be a choice for operator. However from the ubcriber point of view the claifying method could be omewhat clearer due to the fixed threhold value for each traffic load clae. The price of the call could then be predefined in the beginning of the call. The price optionally hown to the ubcriber in the beginning of the call hould remain the ame during the whole call to avoid uncertainty of the total price for the call. V. VOLUME ASED PRICING In the future network connection are planned to be charged by the amount of bit tranferred during call, connection or actually the connection can be called a eion becaue of the imilarity to the computer connection. In G/G network there will be alway open primary PDP context connection, which can be ued for call ignaling and alo to other hared channel tranmiion. It i probable that thi primary connection will be free, cot of uage included in the ubcriber monthly ubcription fee. The econdary PDP context connection will be charged. The charging can be divided into two different categorie depending on the nature of the tranmiion. Real Time connection, Converational and Stream-cla traffic, can be charged by the duration and the volume of bit tranferred during the call. The reaon for thi i that a RT-connection alway occupie a whole traffic channel due to the trict delay requirement of the call. No hared channel can be ued. Non-Real-Time, Interactive and ackground can ue charged channel and there fore even the primary PDP context channel can be hared for thee connection if QoS flow template are compatible. In the G and.g network pricing i till baed on the duration of the call and ervice ued. The only exception i GPRS data traffic, which ue GSM timelot but i charged by the tranferred bit during connection or eion. In the cae that the traffic volume during a connection rie very high there mut be a volume dicount method built in the charging ytem, which prevent charge to rie exceively. Thi can be achieved by uing ame linearity factor, (e e -x* ), for flattening the highet end of the volume pricing curve a in Fig.. Volume pricing formula can be expreed in the ame form a the bandwidth baed pricing,. The main difference in the pure volume baed cae i that the traffic load factor i miing or here it equal (T l % / =), equation 6. P = A * ( e e ) (6) v ij Where: P v = Unit price of the tranferred bit, A ij = Contant unit price for a ubcriber and traffic cla traffic and (e e - x* ) = linearity factor. Volume pricing Traffic volume Fig.. Volume baed pricing VI. SIMULATION RESULTS Mbit Thi chapter preent reult of our imulation. Simulation have been realized by uing the model, which ue different block (call generator, CAC, QoS, pricing, plot) to make pricing and bandwidth allocation deciion. A. optimization optimization can be viewed from two different point of view. Invetment view i the amount of capital, which i inveted into the network, and the income view i the income of the network during a pecified duration of time. The ratio between invetment and the income mut be uch that the baic interet rate of the invetment and the profit expectation will be covered. To be able to calculate the price optimization parameterization to the network there mut be baic knowledge of the ratio between the total price and capacitie (bandwidth, throughput, edge-to-edge delay, ER etc.) of the network. From thee baic factor we are able to calculate the baic unit price for each network reource.. elaticity There mut be alo knowledge of the ubcriber behavior or demand ratio to the price of the network reource. ( elaticity) [Fro]. Eq. 7 how etimation for the price elaticity, (P e ). P e = (Q c /Q) / (P c /P) (7) Where: Q c = change of demand, Q = total, P c = change of price and P = price. The etimated price elaticity will be in the range of -. and -.7 for future G network call and connection. C. Subcriber acceptance and maximizing the total revenue

4 A the price of the call raie the acceptance level of the ervice decreae. Thi behavior can be preented a in Fig..... diagram Fig.. Subcriber acceptance diagram (etimated example) The acceptance (and uage) of a ervice i highly dependent of the pricing of the ervice. When the ervice i cheap all ubcriber needing the ervice will alo ue it and there will be no price-baed limitation of the uage. In the other end there i a different ituation. The price of the ervice i higher than the uefulne of it to the ubcriber and finally when the price i high enough nobody ue the ervice. The optimal price range would be in the middle of the acceptance curve where the ratio of acceptance change and price change are maximized, equation 8. A max (8) P Where: A = F(P ) = factor, P = of the ervice and -denote change. Maximizing the product of P c * Qc max[ P c * Q c ] (9) and taking the correponding P and Q value we get the maximized total revenue equation. max R = A * P * Q () Where: R = total revenue for a ervice and Q = Service (d unit of a ervice). There i alo the quetion of the invetment in capacity to provide the ervice, which rie along with the uage of the ervice and it mut be taken to account when etimating the total profitability of the ytem and ervice. elaticity and acceptance factor can alo be tudied together by comparing how much acceptance increae or decreae when the price change. Thi can be expreed with the following equation. A = F(Q*P e *P c /Q c ) () There are two ub optimal area and view how the optimization proce can be done. The firt ub optimal operating area can alo be expreed a the area where the maximum price elaticity factor i achieved. Thi mean that the network load control by pricing i the mot effective. Thi may be deirable in cae where the operator want coniderably controllable demand for their ervice for performance reaon. The econd ub optimum can be found near the maximum acceptance factor. Coming downward the acceptance there can be found a point where the total revenue R will be maximized, eq.. The area above thi point i actually wate of network reource and it hould be avoided by keeping the pricing lightly above that point. Keeping thee two ub optimum point in mind the obviou operating area (pricing) would be between thee point. Then the operator have the good (if not exactly maximum) control over the network reource and at the ame time the revenue are near optimized. D. and price elaticity dependency A can bee een in the Fig. 6 the hape of the G ervice acceptance curve actually depend on the price elaticity factor of the ervice. % 8 6 curve and price elaticity dependency example Relative uage of the ervice ( % of max available) elaticity range.7 to., not applicaple in G elaticity range. to.7 applicaple in G Fig. 6.G ervice acceptance and price elaticity dependency example A preented in the Fig. 7 the acceptance curve A preent the cae where the price elaticity change from. to. in tep of./% of relative uage of ervice. Curve preent tep./% and repectively and curve C./% and curve D./% price elaticity factor cae. Thee reult have not been verified in a real G environment and one of the future tudy iue will be certification of thi aumption. Thi mean that when the elaticity factor tep and the actual price elaticity increae the ervice become earlier unattractive to ubcriber. A a concluion we can determine that the lower tep ratio produce more profit lowering the income le per amount of change in price.

5 Elaticity Effect to the Acceptence Curve Continuou pricing % curve A. curve. curve C. curve D. A(x) x A(x) Linearity factor L(x) Aij*L(x)*A(x), Aij= Aij*L(x)*A(x), Aij= Relative uage of the ervice (% of max. available) Fig. 7. Elaticity Effect to the Curve E. Optimized price parameterization For parameterization we have ued the price elaticity model with predict the demand change compared to the price change. y uing correct model for acceptance curve we can predict quite accurately the behavior of the income of the network. How ever there i no tudie yet available for ubcriber behavior in G network o ome aumption have to be made to the ubcriber behavior and the price elaticity factor. F. Ue of Linearity factor for adjuting the acceptance curve There i one intereting poibility to adjut the Linearity Factor to match the acceptance curve a in equation and. L(x)=A ij /F(P )= A ij /A = A ij * (e e -x* ) () And L(x) * A ij / F(P ) = A ij, contant () Thi would mean that the acceptance level could be kept to a certain level while the uage of the ervice i increaing by lowering the price appropriately. Contant A ij would actually be the flat rate cae. Thi ituation may be unacceptable due to the fat growing network reource requirement. y adjuting the price factor A ij can control the load and uage of the ervice a hown in the Figure 8. Thi would alo mean that the reource requirement of the network would increae, a the acceptance of the ervice remain table. To avoid over load and over reerving the network reource the acceptance level hould be kept at the level that the network reource are adequate and there i no over invetment to the network infratructure. Contant factor A ij mean that the price change have been /A ij. The price elaticity, which could be in the range of -. and -.7, ha not been taken into account. Thi mean that a price decreae of % would mean an increae of uage by.9 to. percent. Figure 8. linearization VII. CONCLUSIONS Due to the high network auction price operator are very concerned with the income of their future invetment. Our model can be ued to invetigate the behavior of G/G ytem in variou price cenario. Thee kind of pricing evaluation ytem are needed for operator to be able to predict the income of their invetment in the hort and long term time cale. VIII. REFERENCES [] J. F. MacKie-Maon and H. Varian, Pricing congetible network reource, IEEE J. Select. Area Commun, vol, pp. -9, Sept 99 [] H. Jiang, S. Jordan, A pricing model for high peed network with guaranteed quality of ervice, Proc. Conf. Computer Communication "IEEE Infocom, San Francico, CA, Mar 996 [] S. Low, P. Varaiya, An algorithm for optimal ervice proviioning uing reource pricing, Proc. Conf. Computer Commun. IEEE Infocom, Toronto, Canada 99 [] H. Ji, J. Y. Hui, E. Karaan, Go-baed pricing and reource allocation for multimedia broadband network, In proceeding of the IEEE Infocom, page -7, 996 [] Che, Yeon-Koo, Deign competition through multidimenional auction, Rand Journal of Economic, 99 [6] M. ichler, M. Kaukal, A. Segev, Multi-attribute auction for electronic procurement, In Proceeding of the Firt IM IAC workhop on Internet baed Negotiation Technologie, Yorktown Height, NY March 999 [7] Timo Hamalainen, Eero Walleniu, Cae tudy for G/UMTS ervice and billing method, ITC 7, Salvador da ahia, razil, December [8] QoS concept and architecture, GPP TS.7 V.,