1: \documentclass[12pt,fleqn]{article}
2: \usepackage{graphicx}
3: \usepackage{ifthen}
4: \newtheorem{law}{Law}
5: \newtheorem{claim}{Claim}
6: \newtheorem{definition}{Definition}
7: \title{Classes of service under perfect competition and technological change:
8: a model for the dynamics of the Internet?}
9: \author{Daniel Lehmann\thanks{
10: School of Engineering and Computer Science, Hebrew University,
11: Jerusalem 91904, Israel.
12: e-mail: lehmann@cs.huji.ac.il}
13: }
14:
15: \begin{document}
16: \parindent0.0cm
17: \maketitle
18:
19: \newboolean{color}
20: %\setboolean{color}{true}
21: \setboolean{color}{false}
22:
23: \begin{abstract}
24: Certain services may be provided in a continuous, one-dimensional,
25: ordered range of different qualities
26: and a customer requiring a service of quality $q$
27: can only be offered a quality superior or equal to $q$.
28: Only a discrete set of different qualities will be offered,
29: and a service provider will provide the same service (of fixed quality $b$)
30: to all customers requesting qualities of service inferior or equal to $b$.
31: Assuming all services (of quality $b$) are priced identically,
32: a monopolist will choose the qualities of service and the prices
33: that maximize profit but, under perfect competition,
34: a service provider will choose the (inferior) quality of service
35: that can be priced at the lowest price.
36: Assuming significant economies of scale,
37: two fundamentally different regimes are possible:
38: either a number of different classes of service are offered (DC regime),
39: or a unique class of service offers
40: an unbounded quality of service (UC regime).
41: The DC regime appears in one of two sub-regimes: one, BDC,
42: in which a finite number of classes is offered,
43: the qualities of service offered are bounded
44: and requests for high-quality services are not met,
45: or UDC in which an infinite number of classes of service are offered
46: and every request is met.
47: The types of the demand curve and of the economies of scale,
48: and not the pace of technological change,
49: determine the regime and the class boundaries.
50: The price structure in the DC regime obeys very general laws.
51: \end{abstract}
52: \section{Introduction}
53: \label{sec:intro}
54: % Continuous range
55: % Formation of classes of service: tradeoff between large number of users
56: % and quality of service given and cost
57: % Economies of scale
58: % Dynamics
59: % Competition
60: % Question: Structure of the classes and the prices
61: % Statics and Dynamics
62: % Main results: number of differentiated classes catering to the lower
63: % end, the upper end of the market may not be served, but new classes
64: % appear at a constant rate
65: % the classes boundaries are fixed
66: % the prices fall rapidly, in each class at a different
67: % time
68: % the price ratio between a class and the class below it falls rapidly
69: % and converge to a fixed ratio between 1.5 and 3
70: % traffic goes up exponentially
71: % revenue goes up and stabilizes
72: % when demand for quality services is very strong and economies of scale
73: % extremely significant, the fat dumb pipe model will prevail: this has
74: % nothing to do with the pace of change.
75: \subsection{Background}
76: \label{subsec:back}
77: Consider a delivery service.
78: The quality of the service given may be measured
79: by the delay with which the object is delivered.
80: A service guaranteeing same day delivery is of higher quality
81: than one that guarantees only next day delivery, or delivery within
82: three-business days.
83: Assuming that the service can be provided on a continuous range of qualities,
84: it is typically unrealistic to assume that a service provider could provide
85: all those different services to the customers requiring them or bill them
86: differentially for the qualities requested. In many situations, the service
87: provider will have to decide on certain discrete qualities, i.e., classes
88: of service, and will provide only services of those specific qualities.
89:
90: The price paid by the different customers depends only on the quality
91: of the service they receive, not on the (inferior) quality they requested
92: and all customers getting the same service pay the same price.
93: We assume that every customer has sharp requirements concerning the
94: quality of service he requests: he will under no circumstances accept
95: a quality that is inferior to his requirements, even if this is much cheaper,
96: and he will, as long as the price is agreeable, use the service of the
97: least quality that is superior or equal to his request.
98: This assumption is not usual, and not in line with~\cite{GibbensMasonStein:99}
99: for example. Nevertheless, it is very reasonable in connection with Internet
100: services: there is quite a sharp boundary between delays, jitters and latencies
101: that allow for the streaming of video content and those that do not.
102:
103: We assume significant economies of scale to the service provider:
104: the cost of providing $w$ services grows less than linearly with $w$,
105: but we also assume that services of different qualities do not aggregate
106: to generate economies of scale: the cost of providing two classes of service
107: is the sum of the costs of those two different services.
108: This is obviously a severe assumption.
109: In practice, a company offering a few different
110: qualities of service will benefit from economies of scale across those
111: different services: administrative services for example will be shared.
112: But if one considers only the cost of moving packets over the Internet
113: and one assumes, in no way a necessary assumption but a definite possibility
114: (see~\cite{Odlyzko:PMPEC99}), that separate sub-networks are affected
115: to the different qualities of service, then the assumption will be
116: satisfied.
117: In the sequel we may therefore assume that a firm provides only a single
118: quality of service: different firms may provide different qualities of service.
119: In deciding what quality or qualities of service to provide,
120: a firm has to avoid two pitfalls:
121: offering a service of poor quality may cater to a share of the market
122: that is too small to be profitable,
123: offering a service of high quality may involve costs that are too
124: high to attract enough customers.
125: A monopoly will set a quality of service and a price that maximize its
126: revenues, but,
127: in a competitive environment, a customer will buy from the provider offering
128: the lowest price for a service whose quality is superior or equal to
129: the quality requested. This will, in general, result in a lower quality
130: of service, a lower price and a higher activity.
131:
132: Will more than one class of service be proposed?
133: How many? What will the price structure of those classes be?
134: The relation between the traffics in those different classes?
135: Between the revenues gathered in giving those services of different
136: qualities?
137:
138: Those questions may be asked in many different situations, but they are
139: particularly relevant in connection to the Internet.
140: The Internet delivers packets of information from a source to a destination.
141: The IPv4 protocol, the current protocol for the Internet,
142: reserves three bits in each packet for specifying the quality of service
143: desired, but does not use those bits and treats all packets equally.
144: End users rarely pay per packet and usually pay a flat rate.
145: A number of companies have prepared products for QoS (quality of service),
146: i.e., for controlling an Internet-like network in which packets are treated
147: differentially, but the need for such products is not yet proven.
148: A survey of the different proposals for QoS may be found in~\cite{FergHust:98}.
149: Since the Internet is a loose organization that is the product of cooperation
150: between very diverse bodies, billing for such services would also be a major
151: problem.
152: Many networking experts have therefore claimed that the current
153: {\em fat dumb pipe}
154: model is best and argued that it will prevail due to the rapid decline
155: in the cost of equipment.
156: A discussion of the different
157: predictions about the evolution of the Internet may be found
158: in~\cite{Odlyzko:PMPEC99}, where a specific proposal, Paris Metro Pricing
159: (PMP) is advocated.
160: Much interesting information about the economics of the Internet is found
161: at~\cite{Varian:website}.
162: An important aspect of the economy of the Internet is that the
163: prices of the networking equipment are dropping very rapidly. One wonders
164: about the consequences of this rapid decline on the price structure.
165: \subsection{Main results}
166: \label{subsec:mainresults}
167: Answers to the questions above are obtained, assuming a one-dimensional
168: ordered continuum of qualities of service,
169: significant economies of scale and a competitive environment.
170: Under some reasonable and quite general assumptions about the demand curve
171: and the cost function,
172: and assuming that a change in price has a similar effect on the demand
173: for all qualities of service, it is shown that one obtains
174: one of two situations.
175: \begin{itemize}
176: \item (UC regime) If the demand for services of high-quality is strong
177: and the economies of scale are substantive, then a service provider may
178: satisfy requests for service of arbitrary quality: there will be only one class
179: of service catering for everybody's needs; this is the {\em fat dumb pipe}
180: model.
181: \item (DC regime) In other cases, a number of classes of service
182: will be proposed and priced according to the quality of the service provided.
183: Very high quality services may not be provided at all.
184: \end{itemize}
185:
186: The DC regime just described may appear in one of two sub-regimes:
187: \begin{itemize}
188: \item (BDC regime) A finite number of classes of service are offered
189: and services of very high quality are not offered at all.
190: \item (UDC regime) An infinite number of classes and all qualities of service
191: are offered.
192: \end{itemize}
193: Notice that, for example, it cannot be the case that a finite number
194: larger than one of classes are offered and that the highest quality
195: of service caters for unbounded qualities.
196: Significant economies of scale and high demand for high-quality services
197: imply a UC regime, independently of the way prices influence demand.
198: When not in a UC regime, high sensitivity of demand to price implies
199: a BDC regime whereas low sensitivity implies an UDC regime.
200: In a BDC regime, when the basic price of the equipment drops,
201: new classes of service will
202: be offered to the high-end customers that could not be profitably catered
203: for previously.
204: In the UC regime, a slowing of the decline
205: in the price of equipment, does not cause the appearance of multiple classes
206: of service. In a DC regime,
207: a change in the price of equipment cannot cause a transition to
208: a UC regime.
209:
210: A decline in equipment prices, in a DC regime, does not change
211: the boundaries between the different classes of service,
212: but new classes catering to the high-end of the market become available
213: when prices are low enough to make them profitable.
214: The prices drop for all classes, but they drop dramatically for the
215: newly created classes. The ratio of the prices between a class of service
216: and the class just below it decreases and tends to some number typically
217: between $1.5$ and $3$, depending on the exact shape of the distribution
218: of demand over different qualities and the size of the economies of scale,
219: when prices approach zero.
220: Traffic increases when equipment prices decrease.
221: The direction in which revenues change depends on the circumstances,
222: but, typically, revenues go up and then down when equipment prices decrease.
223: The traffics in neighboring classes of service tend to a fixed ratio
224: when prices approach zero.
225:
226: \section{Previous work}
227: \label{sec:previous}
228: % Two types Fishburn and Odlyzko
229: This work took its initial inspiration in the model proposed by A. Odlyzko
230: in the Appendix to~\cite{Odlyzko:PMPEC99}.
231: In~\cite{Odlyzko:PMPEC99,FishOdly:98}, the authors study the case of two
232: different types of customers each type requesting a specific quality of
233: service. They show that, in certain cases, two classes of service will
234: be proposed.
235: The present work assumes a continuous distribution of types of customers.
236: The model of~\cite{GibbensMasonStein:99} is more detailed than the one
237: presented here and addresses slightly different questions, but its conclusions
238: are supported by the present work.
239:
240: The situation described in this paper may seem closely related to
241: the well-studied phenomenon of non-linear pricing~\cite{Wilson:Nonlinear}.
242: A second look seems to indicate major differences:
243: \begin{itemize}
244: \item here the pricing is linear: a customer given many services
245: pays for each of those separately and, in the end, pays the sum of the prices
246: of the services received,
247: \item the cost of providing services is assumed to
248: exhibit significant economies of scale, whereas costs are assumed
249: to be additively separable among customers and qualities
250: in most works on non-linear pricing,
251: \item the present work deals with a competitive environment whereas many
252: works on non-linear pricing assume a monopoly situation.
253: \end{itemize}
254: \section{Model and assumptions}
255: \label{sec:model}
256: % Types
257: % Cost: The technological constant
258: % Demand curve
259: % First Order Approx
260: % Equation
261: % Solution
262: The results to be presented are very general, i.e.,
263: they hold under very mild hypotheses concerning the form of the cost function
264: and the demand curve, and this is the main reason for our interest in them.
265: Nevertheless, describing exactly the weakest hypotheses needed for
266: the results to hold would lead to a paper loaded with mathematical
267: details distracting the reader from the essentials.
268: Therefore, the conditions under which those general results
269: hold will not be described precisely
270: and those results will be only justified by hand waving.
271: Precise results will deal with two specific parametrized families of functions
272: for cost and demand, described in~(\ref{eq:paramf})
273: and~(\ref{eq:paramh}).
274:
275: We consider a service that can be given at a continuum of different
276: levels of quality, e.g., the acceptable delay in delivering a packet
277: over the Internet.
278: In this paper, quality is assumed to be one-dimensional (and totally
279: ordered). This is a severe assumption. Quality of service over the
280: Internet is generally considered to be described by a three-dimensional
281: vector: delay, latency and jitter.
282: In this work the quality of service is characterized by
283: a real number \mbox{$q \geq 1$}. A service quality of $1$ is the lowest
284: quality available.
285: How does a service of quality $q$ compare with a service of quality $1$?
286: The following will serve as a quantitative definition of the quality $q$
287: of a service:
288: \begin{quotation}
289: {\bf Quality} The cost of providing a service of quality $q$
290: is the cost of providing $q$ services of quality $1$.
291: \end{quotation}
292: We assume significant economies of scale: the cost of providing
293: $w$ services of quality $1$ grows like \mbox{$w^{s}$} for some $s$,
294: \mbox{$0 < s < 1$}:
295: the smaller the value of $s$, the larger the economies of scale.
296: In~\cite{FishOdly:98} one may find a discussion
297: of which $s$ best fits the Internet data: values between $1/2$ and $3/4$
298: seem reasonable.
299: The cost of providing $w$ services of quality $1$ is therefore:
300: \[
301: {cost} = c \: {w}^{s}
302: \]
303: where $c$ is some positive value that characterizes
304: the price of the equipment. The number $c$ represents
305: the level of the technology, it is a {\em technological constant}.
306: Many studies indicate that the price of computer and Internet equipment is
307: decreasing at a rapid pace, probably exponentially. The dynamics of our
308: model is described by a decrease in the value of $c$.
309:
310: The cost of providing $f(q)$ services of quality $q$ for every $q$
311: is:
312: \[
313: {cost} = c \: \left( \int_{1}^{\infty} \! q \: f(q) \: dq \right) ^ s.
314: \]
315: As is customary, we include the profits in the costs and will
316: describe the equilibrium by an equation equating costs and revenues.
317: Note that the equation above expresses the fact that economies of scale
318: are obtained only among services of the same quality and do not obtain
319: for performing services of two different qualities.
320:
321: The second half of our model consists of a demand curve describing
322: the demand for services of quality $q$ at price $p$.
323: We assume that the demand for such services is described by a density
324: function \mbox{$d(q,p)$}.
325: The demand for services of quality
326: between $a$ and $b > a$ is:
327: \[
328: {demand} = \int_{a}^{b} \! d(q,p) \: dq
329: \]
330: if the price of any such service is $p$.
331: Notice that we assume that services of different quality are priced
332: identically.
333: It is very difficult to come up with justifiable assumptions about
334: the form of the demand function \mbox{$d(q,p)$}.
335: One may assume that, for any fixed $q$, the value of $d(q,p)$ is
336: decreasing in $p$, and even approaches zero when $p$ tends to infinity.
337: The rate of this decrease is much less clear.
338: For a fixed $p$, how should $d(q,p)$ vary with $q$?
339: Equivalently, what is the distribution of the demand over different qualities
340: of service?
341: This is not clear at all. Both cases of $d(q,p)$ increasing and decreasing
342: in $q$ will be considered.
343: Fortunately, our results are very general and need not assume much about
344: those questions.
345: There are two assumptions we must make, though. They will be described
346: and discussed now.
347: The first one is that the effect of price on the demand is similar at
348: all quality levels, or that the distribution of demand over $q$ is the
349: same at all prices.
350: We assume that the function $d$ is the product
351: of two functions, one that depends only on $q$ and one that depends only
352: on $p$. In the absence of information (difficult to obtain) on the
353: exact form of the demand curve, this sounds like a very reasonable
354: first order approximation.
355: \begin{equation}
356: \label{eq:decoup}
357: {\bf Decoupling} \ \ d(q,p) = f(q) \: h(p)
358: \end{equation}
359: The function $f$ describes the distribution of the demand over different
360: qualities of service.
361: The (decreasing) function $h$ describes the effect of price on the demand.
362: Our results depend on this Decoupling Assumption but should
363: be stable under small deviations from this approximation.
364: The problem of studying systems in which this assumption does not hold
365: is left for future work.
366: Our second assumption is that the function $f$ is constant in time, i.e.,
367: does not depend on the technological constant $c$.
368: Technological progress modifies demand only through price.
369: Our results about the dynamics of the model rely on this assumption.
370:
371: To be able to prove mathematically precise results, we shall, when needed,
372: assume that:
373: \begin{equation}
374: \label{eq:paramf}
375: f(q) = q^{\alpha}
376: \end{equation}
377: for some $\alpha$:
378: the larger the value of $\alpha$, the larger the relative size of the demand
379: for high-quality services.
380: Is it clear whether $\alpha$ should be
381: positive or negative? Suppose e-mail packets and streaming-video packets
382: can be sent at the same price, would the demand for streaming-video packets
383: be larger than that for e-mail? Probably. It seems that a positive
384: value for $\alpha$ is more realistic, but
385: we do {\em not} make any assumption on the sign of $\alpha$.
386: It may be the case that a more realistic function $f$ should, first,
387: for $q$'s close to one, increase, and then decrease for $q$'s above the
388: quality required for the most demanding applications at hand.
389: We shall not try to discuss this case here, but the conclusions of this paper
390: offer many qualitative answers even for such a case.
391:
392: For $h$, we shall always assume that
393: \mbox{$h(p) \geq 0$} for every \mbox{$p \geq 0$}, that $h$ is decreasing
394: when $p$ increases and that, when $p$ approaches $+ \infty$,
395: $h(p)$ approaches zero.
396: Since $h(0)$ characterizes the demand at price $0$, we shall also assume,
397: for proper scaling, that \mbox{$h(0) = 1$}.
398: The following are examples of possible forms for $h$:
399: \begin{equation}
400: \label{eq:paramh}
401: h(p) = \frac{1}{1 + {(a \: p)}^{\beta}}, {\rm \ for \ some \ } \beta > 0, \
402: a > 0
403: \end{equation}
404: In~(\ref{eq:paramh}), $\beta$ describes the asymptotic rate of decrease
405: of the demand when price increases, but $a$ describes the transient behavior of
406: the demand:
407: the larger $\beta$, the more sensitive the demand is to price.
408: For $\beta$ close to zero, the demand is little influenced by price, i.e.,
409: the demand is incompressible.
410: One may also consider the following:
411: \begin{equation}
412: \label{eq:exp}
413: h(p) = e^{-p},
414: \end{equation}
415: \begin{equation}
416: \label{eq:expsq}
417: h(p) = e^{-p^2}.
418: \end{equation}
419: As explained at the start of this Section, precise mathematical
420: results will assume that the functions $f$ and $h$ are defined
421: by~(\ref{eq:paramf}) and~(\ref{eq:paramh}), but the qualitative results
422: hold for a much larger family of functions.
423: \section{Beginnings}
424: \label{sec:begin}
425: \subsection{Equilibrium Equation}
426: \label{subsec:equilibrium}
427: As explained in Section~\ref{sec:intro}, we assume that the service provider
428: cannot provide differentiated services: it has to provide
429: the same service and to charge the same price for every service it accepts
430: to perform.
431: Let us, first, assume that there is no service available and that
432: a service provider considers whether or not to enter the market.
433: In a sense, the provider is a monopolist for now.
434: But we shall show below that, because of possible competition, it cannot
435: maximize its profit.
436: It is clear that a provider will choose a quality
437: of service $b$ and a price $p$, and offer services of quality $b$
438: at price $p$ to any potential customer.
439: The potential customers are all those customers who request services of quality
440: \mbox{$q \leq b$}.
441: Customers requesting quality above $b$ will not be served.
442: If there are no competitors, one can hope to attract
443: the totality of the demand for services of quality \mbox{$q \leq b$}
444: at price $p$.
445: The size of the demand for service of quality $q$ (\mbox{$q \leq b$})
446: is \mbox{$f(q) h(p)$}. Since we provide a service of uniform quality
447: $b$, the equivalent number of services of quality $1$ that we shall
448: perform is:
449: \begin{equation}
450: \label{eq:X}
451: \int_{1}^{b} \! b \: f(q) \: h(p) \: dq = b \: h(p) \: A(1,b),
452: \end{equation}
453: where we define:
454: \[
455: A(a,a') = \int_{a}^{a'} \! f(q) \: dq.
456: \]
457: Note that $A(q,q')$ is the demand for services of quality between $q$
458: and $q'$ if the price for those services is zero.
459: The cost of providing the services of~(\ref{eq:X}) is:
460: \begin{equation}
461: \label{eq:cost}
462: {cost} = c \: {b}^{s} \: h^{s}(p) \: A^{s}(1,b).
463: \end{equation}
464: The total demand for services is:
465: \begin{equation}
466: \label{eq:demand}
467: {demand} = \int_{1}^{b} \! f(q) \: h(p) \: dq = h(p) \: A(1,b).
468: \end{equation}
469: Since every service pays the same price $p$, the revenue is:
470: \begin{equation}
471: \label{eq:revenue}
472: {revenue} = p \: h(p) \: A(1,b).
473: \end{equation}
474: Equilibrium is characterized by:
475: \[
476: p \: h(p) \: A(1,b) = c \: {b}^{s} \: h^{s}(p) \: A^{s}(1,b),
477: \]
478: or equivalently by:
479: \begin{equation}
480: \label{eq:eq}
481: p \: h^{1 - s}(p) = c \: {b}^{s} \: A^{-(1-s)}(1,b).
482: \end{equation}
483: Equation~(\ref{eq:eq}) characterizes the price $p$ as a function of the quality
484: of service $b$. Most of the remainder of this paper is devoted to the study
485: of this equation.
486: % Explain now why the lhs has a unique maximum
487: % No solution or two solutions, in which case the low price is taken
488: % Different $b$'s provide different prices, revenues and traffics
489:
490: \subsection{Existence of an equilibrium}
491: \label{subsec:price}
492: Given $b$ and $c$, we want to solve~(\ref{eq:eq}) in the variable
493: $p$.
494: The solutions of~(\ref{eq:eq}) are obtained as the intersections
495: of the curve representing the left hand side and a horizontal line
496: representing the right hand side.
497: The left hand side is equal to zero for \mbox{$p=0$} and is always
498: non-negative.
499: It is reasonable to assume that, considered as
500: a function of $p$, it has one of the two
501: following forms.
502: \begin{figure}[htbp]
503: \centering
504: \ifthenelse{\boolean{color}}
505: {\includegraphics[height=8cm]{lhs1col.ps}}
506: {\includegraphics[height=8cm]{lhs1.ps}}
507: \caption{Left hand side of~(\ref{eq:eq}), $s = 2 / 3$, $\beta = 6$}
508: \label{fig:lhs1}
509: \end{figure}
510: \begin{itemize}
511: \item Sensitive case: \mbox{$p h^{1-s}(p)$} increases, has a maximum and
512: then decreases. This seems to be the most typical case: the price has
513: a strong influence on the demand and therefore $h$ is such that
514: \mbox{$h^{1 - s}(p)$} decreases more rapidly than $1 / p$ when $p$ is large.
515: This is the case, for any $s$,
516: when $h$ has the form described in~(\ref{eq:exp})
517: or~(\ref{eq:expsq}). If $h$ follows~(\ref{eq:paramh}), it is the case
518: iff \mbox{$\beta > \frac{1}{1-s}$}.
519: Fig~\ref{fig:lhs1} presents a typical sensitive case:
520: \mbox{$s = 2 / 3$}, \mbox{$\beta = 6$} and \mbox{$a = 1$}, i.e.,
521: \mbox{$h(p) = \frac{1}{1 + p^{6}}$}.
522: \item Insensitive case: \mbox{$p h^{1-s}(p)$}
523: increases always and tends to
524: \mbox{$+ \infty$} when $p$ tends to \mbox{$+ \infty$}.
525: This is the case if the demand is relatively insensitive to the price.
526: It happens if $h$ follows~(\ref{eq:paramh})
527: and \mbox{$\beta < \frac{1}{1-s}$}.
528: The case in which \mbox{$p h^{1-s}(p)$} increases and approaches a finite
529: value when $p$ tends to \mbox{$+ \infty$} is possible
530: (\mbox{$\beta = \frac{1}{1-s}$}) but not generic.
531: Fig~\ref{fig:lhs2} presents a typical insensitive case:
532: \mbox{$s = 2 / 3$}, \mbox{$\beta = 2$} and \mbox{$a = 1$}, i.e.,
533: \mbox{$h(p) = \frac{1}{1 + p^{2}}$}.
534: \end{itemize}
535: \begin{figure}[htbp]
536: \centering
537: \ifthenelse{\boolean{color}}
538: {\includegraphics[height=8cm]{lhs2col.ps}}
539: {\includegraphics[height=8cm]{lhs2.ps}}
540: \caption{Left hand side of~(\ref{eq:eq}), $s = 2 / 3$, $\beta = 2$}
541: \label{fig:lhs2}
542: \end{figure}
543: Equation~(\ref{eq:eq}) behaves quite differently in those two cases.
544: Whereas in the insensitive case, (\ref{eq:eq}) has a unique solution
545: whatever the values of $c$ and $b$, in the sensitive case
546: we must distinguish between two situations.
547: \begin{itemize}
548: \item If the horizontal line is high, e.g., if the technological constant $c$
549: is large, the horizontal line does not intersect the curve and
550: there is no price $p$ to solve~(\ref{eq:eq}). In this case
551: providing a service of quality $b$ is unprofitable, even in the absence
552: of competition and no such service will be provided.
553: \item If the horizontal line representing the right hand side intersects
554: the curve, for each intersection there is a price
555: $p$ solving~(\ref{eq:eq}). In this case,
556: the provider may consider providing a service of quality $b$ at any such
557: price $p$, but before any more sophisticated analysis, it is clear that
558: the price at which service of quality $b$ will be proposed (if proposed)
559: is the smallest of the $p$'s solving~(\ref{eq:eq}).
560: If a provider proposes another solution of the equation, a competitor
561: will come in and propose a service of the same quality at a lower price.
562: \end{itemize}
563: In other terms: if~(\ref{eq:eq}) has no solution, no service of
564: quality $b$ can be profitably provided, and if~(\ref{eq:eq}) has
565: at least a solution then the service may be provided at a price that is
566: the smallest solution of~(\ref{eq:eq}).
567:
568: \subsection{Quality of the service provided in a competitive environment}
569: \label{subsec:service}
570: Suppose a service provider finds it is in a situation
571: in which~(\ref{eq:eq}) has solutions for some values of $b$.
572: It may consider providing a service of quality $q$ for any one of those
573: values. The price it will charge, the traffic and the revenue obtained
574: depend on the quality $q$ chosen. To understand the choice that
575: a provider has to make, we need to study~(\ref{eq:eq})
576: in some more depth.
577: Assume that $c$ is given. One would like to know which qualities
578: of service can be profitably provided, i.e.,
579: one wants to consider the set of
580: $b$'s for which the Equation has a solution.
581: The left hand side of the Equation has been studied in
582: Section~\ref{subsec:price}: let us study the right hand side.
583: Under very general circumstances, for \mbox{$b \rightarrow {1}^{+}$},
584: \mbox{$A^{-(1-s)}(1,b)$} approaches $+ \infty$.
585: Figure~\ref{fig:rhs} describes the behavior of the
586: right hand side of~(\ref{eq:eq}) as a function of $b$ for different
587: choices of $f$ (i.e. $\alpha$) and $s$.
588: The number $c$ is chosen to be $1$ in all cases.
589: In the leftmost graph \mbox{$\alpha = 1$} and in the rightmost graph
590: \mbox{$\alpha = -1$}.
591: \begin{figure}[htbp]
592: \begin{tabular}{cc}
593: \ifthenelse{\boolean{color}}
594: {\includegraphics[height=6cm]{rhs1col.ps} &
595: \includegraphics[height=6cm]{rhs2col.ps}}
596: {\includegraphics[height=6cm]{rhs1.ps} &
597: \includegraphics[height=6cm]{rhs2.ps}}
598: \end{tabular}
599: \caption{Right hand side of~(\ref{eq:eq})}
600: \label{fig:rhs}
601: \end{figure}
602: The function \mbox{$b^{s} A^{-(1-s)}(1,b)$} also approaches $+ \infty$
603: for $b$ approaching $1$ and, essentially, it may present any one of
604: two possible behaviors: either it decreases always, or it decreases
605: until it gets to a minimum and then increases without bounds.
606: The case it decreases to a minimum and then increases and approaches
607: a finite value when $b$ tends to \mbox{$+ \infty$} is possible but not
608: generic.
609: These two possible behaviors for the function
610: \mbox{${b}^{s} \: A^{-(1-s)}(1,b)$} determine two very different regimes.
611: \begin{definition}
612: \label{def:regime}
613: In the first regime, the Universal Class regime (UC regime),
614: the right hand side of~(\ref{eq:eq})
615: is a decreasing function of $b$. In this regime the set of $b$'s
616: for which~(\ref{eq:eq}) has a solution is an interval of the form
617: \mbox{$[q_{1} , + \infty[$}, for some \mbox{$q_{1} > 1$}.
618: Any service of quality at least $q_{1}$ may be profitably provided.
619: In the second regime, the Differentiated Classes regime (DC regime),
620: the right hand
621: side of~(\ref{eq:eq}) decreases, reaches a minimum and increases.
622: One must then distinguish between the sensitive and the insensitive cases.
623: In the sensitive case, the set of $b$'s for which~(\ref{eq:eq})
624: has a solution is either empty or an interval of the form
625: \mbox{$[q_{1} , q_{2}]$}, for some \mbox{$q_{2} \geq q_{1} > 1$}.
626: This regime (DC, sensitive case) will be called the BDC (bounded DC)
627: sub-regime.
628: In the insensitive case, (\ref{eq:eq}) has a solution for any $b$
629: in the interval \mbox{$[1 , + \infty [$}.
630: This regime (DC, insensitive case) will be called the UDC (unbounded DC)
631: sub-regime.
632: \end{definition}
633: Suppose a service provider finds it is in a situation in which~(\ref{eq:eq})
634: has solutions for some interval of qualities.
635: We know it should choose a quality $q$ in this interval,
636: offer a service of quality $q$ and propose this service at the price $p$
637: that is the smallest solution to~(\ref{eq:eq}).
638: Which $q$ should our provider choose?
639: To each choice for $q$ corresponds a price $p_{q}$.
640: The number of services performed is given by
641: \[
642: {demand} = h(p_{q}) \: A(1,q).
643: \]
644: Since every service performed is a service of quality $q$,
645: and every such service is equivalent to $q$ services of quality $1$,
646: the weighted traffic is:
647: \[
648: {traffic} = q \: h(p_{q}) \: A(1,q).
649: \]
650: The revenue is:
651: \[
652: {revenue} = p_{q} \: h(p_{q}) \: A(1,q).
653: \]
654:
655: Consider, first, a case typical of the DC regime.
656: In Figure~\ref{fig:ptraffrevDC} the price $p_{q}$, the traffic
657: and the revenue are described as functions
658: of the quality $q$ of the service chosen by the provider.
659: The service can be provided only for a bounded interval of qualities.
660: Notice that, in this interval, the price decreases rapidly
661: and then increases slowly,
662: that traffic and revenue increase and then decrease, and that the
663: minimum price is obtained for a quality inferior to the quality
664: that maximizes the revenue.
665: \begin{figure}[htbp]
666: \centering
667: \ifthenelse{\boolean{color}}
668: {\includegraphics[height=8cm]{ptraffrevdccol.ps}}
669: {\includegraphics[height=8cm]{ptraffrevdc.ps}}
670: \caption{Price, Traffic, Revenue as functions of $q$ in the DC regime.
671: Service can be provided only for $q$ in a bounded interval}
672: \label{fig:ptraffrevDC}
673: \end{figure}
674:
675: A case typical of the UC regime is described in
676: Figure~\ref{fig:ptraffrevUC}.
677: By our discussion of~(\ref{eq:eq}), it is clear that $p_{q}$ is
678: a decreasing function of $q$. Notice that traffic and revenue increase
679: with $q$.
680: \begin{figure}[htbp]
681: \centering
682: \ifthenelse{\boolean{color}}
683: {\includegraphics[height=8cm]{ptraffrevuccol.ps}}
684: {\includegraphics[height=8cm]{ptraffrevuc.ps}}
685: \caption{Price, Traffic, Revenue as functions of $q$ in the UC regime}
686: \label{fig:ptraffrevUC}
687: \end{figure}
688:
689: What is the quality of service most profitable?
690: Since the revenue includes the profits, one may assume, at first sight,
691: that the first provider to propose such a service
692: will choose the quality $q$ that maximizes its revenue.
693: This analysis is clearly incorrect in a competitive situation.
694: If a provider decides to offer the service that maximizes its revenue,
695: a competitor provider may come in and offer a service of lower quality
696: at a lower price and steal the demand for the lower quality services.
697: Therefore, the price at which the service will be offered is the
698: minimum price possible: the minimum of the function $p_{q}$ on its interval of
699: definition.
700: \begin{claim}
701: The quality $q$ of the service provided is the $q$ for which $p_{q}$
702: is minimal.
703: \end{claim}
704:
705: This remark turns out to be of fundamental importance because
706: this quality $q$, contrary to the quality that maximizes revenue,
707: does not depend on the technological constant $c$ or on the function $h$
708: that describes the dependence of the demand on the price charged.
709: It depends only on the function $f$ that describes the distribution
710: of the demand over the different qualities of service and on the size
711: of the economies of scale described by the number $s$.
712: Indeed, this quality $q$ is the value of $b$ that minimizes
713: \mbox{$b^{s} A^{-(1-s)}(1,b)$} in the DC regime and it is $+ \infty$ in
714: the UC regime.
715: There is, therefore, a sharp distinction between the two regimes:
716: in the DC regime a service of finite quality is provided,
717: whereas, in the UC regime, the quality of service provided
718: is high enough to please every customer, i.e., unbounded.
719: \begin{law}[First Law]
720: Whether or not some service is provided does not depend on the function
721: $h$ that describes the way demand depends on price.
722: It depends on the technological constant $c$, the size of the economies
723: of scale $s$, and on the function $f$ that describes the distribution of
724: demand over different qualities of service.
725: If some service is provided its quality depends only on $s$ and $f$,
726: it does not depend on $c$ or $h$.
727: In the UC regime, a single class serves requests for services of arbitrary
728: quality and it is given for free.
729: In the DC regime, the service of lowest quality has a finite quality
730: and does not cater for high-end customers.
731: The prevailing regime depends on the form of the economies
732: of scale and on the form of the distribution of the demand over different
733: qualities, it does not depend on the price of the equipment or on the
734: way prices influence demand.
735: \end{law}
736: In Section~\ref{sec:border}, we shall, first, mark the boundary between
737: these two regimes. Then, both regimes: UC and DC will be studied in turn.
738:
739: \section{Between the two regimes}
740: \label{sec:border}
741: In this Section, we shall assume that $f$ has the form described
742: in~(\ref{eq:paramf}), for some $\alpha$, and study the exact boundaries
743: of the UC and DC regimes.
744: The regime is determined by the behavior of the function:
745: \[
746: w(b) = b^{s} \: \left( \int_{1}^{b} \frac {dq}{q^{\alpha}} \right)^{-(1-s)},
747: \]
748: for \mbox{$b \geq 1$}.
749: If $w(b)$ has a minimum, then the regime is DC, if it always decreases,
750: the regime is UC.
751: Three cases must be distinguished.
752: \[
753: {\rm If \ } \alpha < -1 , \ \
754: w(b) = \frac {b^{s}} {{ \left( -\alpha - 1 \right) }^{-(1-s)}} \;
755: { \left( 1 - b^{\alpha+1} \right) }^{-(1-s)}.
756: \]
757: \[
758: {\rm If \ } \alpha = -1 , \ \
759: w(b) = b^{s} \: \ln^{-(1-s)} b,
760: \]
761: and
762: \[
763: {\rm if \ } \alpha > -1 , \ \
764: w(b) = \frac {b^{s}} {{ \left( \alpha + 1 \right) }^{-(1-s)}} \;
765: { \left( b^{\alpha+1} - 1 \right) }^{-(1-s)}.
766: \]
767: For \mbox{$\alpha < -1$}, we have:
768: \[
769: w'(b) = \frac {b^{-(1-s)} \: {\left( 1 - b^{\alpha+1} \right)}^{-(2-s)} }
770: {{ \left( -\alpha - 1 \right) }^{-(1-s)}}
771: \; \left[ s (1 - b^{\alpha+1}) - (1-s) b (-\alpha-1) b^{\alpha} \right] ,
772: \]
773: \[
774: w'(b) = \frac {b^{-(1-s)} \: {\left( 1 - b^{\alpha+1} \right)}^{-(2-s)} }
775: {{ \left( -\alpha - 1 \right) }^{-(1-s)}}
776: \; \left[ (-(2+\alpha) s + \alpha + 1) \: b^{\alpha+1} + s \right] ,
777: \]
778: which approaches $s$ and is therefore positive for large $b$'s.
779: We conclude that for any \mbox{$\alpha < -1$} the regime is DC.
780:
781: For \mbox{$\alpha = -1$},
782: \[
783: w'(b) = b^{s-1} \: \ln^{-(2-s)} b \:
784: \left[ s \ln b - (1-s) \right],
785: \]
786: which is also positive for large $b$'s.
787: For \mbox{$\alpha = -1$} the regime is DC.
788:
789: For \mbox{$\alpha > -1$}, we have:
790: \[
791: w'(b) = \frac {b^{-(1-s)} \: {\left( b^{\alpha+1} - 1 \right)}^{-(2-s)} }
792: {{ \left( \alpha + 1 \right) }^{-(1-s)}}
793: \; \left[ s (b^{\alpha+1}-1) - (1-s) b (\alpha+1) b^{\alpha} \right]
794: \]
795: \[
796: w'(b) = \frac {b^{-(1-s)} \: {\left( b^{\alpha+1} - 1 \right)}^{-(2-s)} }
797: {{ \left( \alpha + 1 \right) }^{-(1-s)}}
798: \; \left[ ( (2+\alpha) s - \alpha - 1 ) \: b^{\alpha+1} - s \right]
799: \]
800: For large $b$'s, $w'(b)$ has the sign of \mbox{$(2+\alpha) s - \alpha - 1$}.
801: We may now conclude our study.
802: \begin{claim}
803: \label{claim:DCUC}
804: \[
805: {\rm the \ regime \ is \ DC \ iff \ } \alpha \leq -1 {\rm \ or \ }
806: s > \frac{\alpha+1}{\alpha+2},
807: \]
808: \[
809: {\rm the \ regime \ is \ UC \ iff \ } \alpha > -1 {\rm \ and \ }
810: s \leq \frac{\alpha+1}{\alpha+2},
811: \]
812: \end{claim}
813: The boundary between the two regimes can be seen in Figure~\ref{fig:UCDC}.
814: \begin{figure}[htp]
815: \centering
816: \ifthenelse{\boolean{color}}
817: {\includegraphics[height=8cm]{ucdccol.ps}}
818: {\includegraphics[height=8cm]{ucdc.ps}}
819: \caption{Boundary between the UC and DC regimes}
820: \label{fig:UCDC}
821: \end{figure}
822: Informally, if the economies of scale are very substantial (small $s$)
823: and if high quality services are in big demand ($f$ grows rapidly),
824: the UC regime prevails.
825: Otherwise, there is an upper bound to the quality of profitable services.
826: There is no profitable way to provide
827: for a service of arbitrary quality because its cost would
828: deter too many potential customers.
829: \section{The universal class regime}
830: \label{sec:univclass}
831: \subsection{A paradoxical regime}
832: \label{subsec:paradox}
833: Let us study the UC regime. As discussed in Section~\ref{subsec:service},
834: Figure~\ref{fig:ptraffrevUC} and in our first Law, in this regime,
835: the right hand side of~(\ref{eq:eq}) is as close to zero
836: as one wishes if one takes $b$ to be large enough.
837: For any value of the technological constant $c$, even a very high one,
838: it is profitable to provide a service of infinite quality at zero price,
839: and therefore, for any value of the technological constant $c$,
840: a unique class of service of unbounded quality is offered for free.
841: This seems quite a paradoxical situation.
842: The demand (\ref{eq:demand}) and the cost of providing the service
843: (\ref{eq:cost}) are both infinite.
844: The consideration of the revenue is interesting.
845: The revenue (see (\ref{eq:revenue}) and~(\ref{eq:eq})) is infinite,
846: but depends linearly on the technological constant $c$.
847: The revenue drops linearly with the price of the equipment.
848:
849: This regime seems to fit the fat dumb pipe model of the Internet:
850: there are enough
851: resources to give every request a treatment of the highest quality,
852: there is no need to give differentiated services and the price is very low
853: since the demand is very high and the economies of scale very substantive.
854:
855: The dynamics of the UC regime under technological progress, i.e., a
856: drop in the constant $c$, is quite paradoxical.
857: Such a drop does not significantly affect the, already negligible,
858: price of the service, but it affects adversely
859: the (infinite) revenues of the provider.
860:
861: The First Law implies that the boundary between the UC and DC regimes
862: does not depend on the technological constant $c$,
863: therefore the answer to the question whether the fat dumb pipe model
864: prevails or whether a number of classes of service will develop
865: depends only on the relation between the economies of scale and
866: the distribution of demand over different qualities, it does not depend
867: on the price of the equipment.
868: Both regimes are stable under technological progress.
869: \subsection{A more realistic model}
870: \label{subsec:realistic}
871: The paradoxical aspects of the UC regime as described
872: above stem from the consideration of a demand curve that assumes
873: significant demand for services of unbounded quality.
874: It seems more realistic to assume an upper bound to the quality requested
875: from services. A specific example will be given now, to explain what
876: the UC regime may look like in practice.
877: Assume that there is an upper bound $q_{m}$ to the quality of service
878: requested and that the distribution of demand $f$ over different qualities
879: is described by:
880: \begin{equation}
881: \label{eq:piecewise}
882: f(q) = \left\{ \begin{array}{ll}
883: q & \mbox{if $1 \leq q \leq q_{m}$} \\
884: 0 & \mbox{if $q_{m} \leq q$}
885: \end{array}
886: \right.
887: \end{equation}
888: Assume also that \mbox{$s = \frac{2}{3}$} and that
889: \mbox{$h(p) = \frac{1}{1+p^3}$}.
890: It follows from our discussion above that the service provided, if provided,
891: will be of quality $q_{m}$.
892: Therefore the equilibrium equation is:
893: \begin{equation}
894: \label{eq:speceq}
895: \frac{p}{(1+p^3)^{1/3}} = 2^{1/3} \: c \: {q_{m}}^{2/3} \:
896: {\left( {q_{m}}^2 - 1 \right)}^{-1/3}.
897: \end{equation}
898: The left hand side of~(\ref{eq:speceq}) has a maximum value of $1$.
899: For any $c$ such that
900: \[
901: c \geq 2^{-1/3} \: {q_{m}}^{-2/3} \:
902: {\left( {q_{m}}^2 - 1 \right)}^{1/3}
903: \]
904: there is no profitable service.
905: For any $c$ smaller than this value,
906: a provider will choose to provide a service of quality $q_{m}$.
907: The price, traffic and revenue are functions of $c$ and described in
908: Figure~\ref{fig:UCreal}.
909: \begin{figure}[htbp]
910: \centering
911: \ifthenelse{\boolean{color}}
912: {\includegraphics[height=8cm]{ucrealcol.ps}}
913: {\includegraphics[height=8cm]{ucreal.ps}}
914: \caption{Evolution of price, traffic and revenue in the UC regime}
915: \label{fig:UCreal}
916: \end{figure}
917: As soon as some service is provided, it caters for all needs.
918: A unique class of service will still prevail in the presence of
919: a drop in the price of equipment, i.e., a decrease in $c$.
920: Notice that the revenue decreases when the technology constant drops.
921:
922: \section{The differentiated classes regime}
923: \label{sec:multiple}
924: % The sequence of boundaries is independent of the technological constant
925: % Growth in the number of classes and types of services provided
926: % Evolution of prices, and price ratios
927: % Evolution of traffic and traffic ratios
928: Let us study the DC regime now.
929: The best approach is to take a dynamic perspective and consider
930: the evolution of the system under a decrease in the technological
931: constant $c$.
932: \subsection{The BDC sub-regime}
933: \label{subsec:BDCregime}
934: In the first part of this discussion we study the BDC sub-regime,
935: i.e., the {\em sensitive} case described in Section~\ref{subsec:price} in
936: which the left hand side of~(\ref{eq:eq}) increases to a maximum
937: and then decreases.
938: We have seen that for high $c$'s no service is provided.
939: When $c$ falls below a certain value, $c_{0}$,
940: a service of some finite quality $q_{0}$ is provided.
941: From the First Law we know that,
942: even in the face of further declines in $c$, the quality $q_{0}$ of the
943: service provided will not change.
944: Two main questions arise:
945: \begin{itemize}
946: \item how do the price, the traffic and the revenue of the provider of
947: a service of quality $q_{0}$
948: evolve under a further decrease in $c$?
949: \item will services of higher quality be offered? which? when? at what price?
950: \end{itemize}
951: The first question is relatively easily disposed of.
952: Since the right hand side of~(\ref{eq:eq}) is a linear function of $c$,
953: the assumption that the left hand side increases to a maximum implies
954: that the price drops rapidly at first (for $c$ close to $c_{0}$),
955: and then at a slower pace. For the functions $h$ considered,
956: the slope of the left hand side is one at \mbox{$p = 0$} and
957: the price $p$ drops essentially linearly with $c$ and approaches zero
958: when $c$ approaches zero.
959: The demand (or the traffic) varies as \mbox{$h(p)$} and therefore
960: increases very rapidly at first and then approaches some finite value.
961: The revenue varies as \mbox{$p h(p)$} or \mbox{$c h(p)^{s}$}.
962: Typically it may exhibit, first, an increase, for values of $c$
963: at which a drop in the price of the service is over-compensated by an
964: increase in demand and then a decrease. For small $c$'s, the revenue
965: drops linearly with $c$.
966: Figure~\ref{fig:ptrbdc1} describes the evolution of price, traffic and revenue
967: for the lowest class of service, the first to be offered on the market.
968: \begin{figure}[htbp]
969: \centering
970: \ifthenelse{\boolean{color}}
971: {\includegraphics[height=8cm]{ptrbdc1col.ps}}
972: {\includegraphics[height=8cm]{ptrbdc1.ps}}
973: \caption{Evolution of price, traffic and revenue in the BDC regime, lowest class of service for $\alpha = -2.5, s = 2/3, \beta = 6$ and $a = 0.7$}
974: \label{fig:ptrbdc1}
975: \end{figure}
976:
977: The second question is deeper. Since a service of quality $q_{0}$ will
978: be provided at a price we have just studied, it is clear that the
979: only customers who could be interested in another service are those
980: requesting a service of quality superior to $q_{0}$.
981: Will such a service, of quality \mbox{$b \geq q_{0}$} be provided?
982: The equilibrium equation now becomes:
983: \begin{equation}
984: \label{eq:eq2}
985: p \: {h(p)}^{1 - s} = c \: {b}^{s} \: A(q_{0},b)^{-(1-s)}.
986: \end{equation}
987: Notice that~(\ref{eq:eq2}) is very similar to~(\ref{eq:eq}).
988: A service of quality $b$ may be profitably offered iff~(\ref{eq:eq2})
989: has a solution. See Section~\ref{subsec:price} for a discussion:
990: the situation is the same. There is a value $c_{1}$ above which
991: there is no solution and below which there is a solution.
992: Since we are in the sensitive case,
993: the left hand side has a maximum and
994: the price of the service, of quality $b$,
995: if provided, will be the smallest $p$ satisfying~(\ref{eq:eq2}).
996:
997: The discussion of which service will be provided is similar to the
998: discussion found in Section~\ref{subsec:service}.
999: The behavior of the right hand side of~(\ref{eq:eq2})
1000: as a function of $b$ is the same as that of~(\ref{eq:eq}).
1001: We have assumed we are in the DC regime, and therefore there is a finite
1002: value $q_{1}$ for $b$ that minimizes
1003: \mbox{${b}^{s} \: A(q_{0},b)^{-(1-s)}$}.
1004: The service provided will be of a quality $q_{1}$.
1005:
1006: In the case $f$ has the form described in~(\ref{eq:paramf}), the
1007: computation of $q_{1}$ is simplified by a change of variable
1008: \mbox{$ q = q_{0} \: q'$}.
1009: \[
1010: A(q_{0},b) = \int_{q_{0}}^{b} \! f(q) \: dq = \int_{1}^{b / q_{0}} \!
1011: f(q_{0} q') \: q_{0} \: dq' = q_{0}^{\alpha+1} A(1,b/q_{0})
1012: \]
1013: and the equation becomes:
1014: \[
1015: p \: {h(p)}^{1 - s} = c' \: {(b / q_{0})}^{s} \: A(1,b / q_{0})^{-(1-s)},
1016: {\rm \ for \ } c' = c \: q_{0}^{-(\alpha+1)(1-s)+s}.
1017: \]
1018: Notice that
1019: \[
1020: -(\alpha+1)(1-s)+s = (\alpha+2)s - (\alpha+1) > 0
1021: \]
1022: by Claim~\ref{claim:DCUC} since we are in the DC regime and therefore
1023: \mbox{$c' > c$}.
1024: Equation~(\ref{eq:eq2}) is the same as~(\ref{eq:eq}), for a larger
1025: technological constant, after our change of variable.
1026: This implies that the service provided will be of quality $q_{1}$ with
1027: \[
1028: q_{1} / q_{0} = q_{0} , {\rm \ i.e., \ } q_{1} = q_{0}^{2}.
1029: \]
1030: It will appear when
1031: \[
1032: c' = c_{0} , {\rm \ i.e., \ } c_{1} = c_{0} \: q_{0}^{(\alpha+1)(1-s)-s}.
1033: \]
1034: The general dynamic picture is now clear.
1035: Different classes of service appear as the technological constant $c$
1036: decreases. The $n$'th class (\mbox{$n \geq 0$}) appears when
1037: \mbox{$c = c_{0} \: q_{0}^{n(-s+(\alpha+1)(1-s))}$} and it is of quality
1038: \mbox{${q_{0}}^{n}$}. For any $c>0$, only a finite number of qualities of
1039: service are proposed and no service is provided for request of high
1040: quality service. Since, from Section~\ref{sec:border}, we may also
1041: easily compute $q_{0}$ and see that:
1042: \[
1043: {\rm if \ } \alpha < -1 , \ \ \ \ q_{0} =
1044: \left( \frac{s}{(2+\alpha) s - \alpha - 1} \right) ^ {\frac{1}{\alpha+1}},
1045: \]
1046: \[
1047: {\rm if \ } \alpha = -1 , \ \ \ \ q_{0} = {e}^{(1-s)/s},
1048: \]
1049: \[
1050: {\rm if \ } \alpha > -1 , \ \ \ \ q_{0} =
1051: \left( \frac{s}{(2+\alpha) s - \alpha - 1} \right) ^ {\frac{1}{\alpha+1}},
1052: \]
1053: the picture is complete.
1054: Figure~\ref{fig:ptrbdc2} describes the price, traffic and revenue in the second
1055: class of service as functions of the technological constant $c$, for the same
1056: constants as in Figure~\ref{fig:ptrbdc1}.
1057: Compare with Figure~\ref{fig:ptrbdc1} and see that the qualitative evolution
1058: is similar in both classes.
1059: \begin{figure}[htbp]
1060: \centering
1061: \ifthenelse{\boolean{color}}
1062: {\includegraphics[height=8cm]{ptrbdc2col.ps}}
1063: {\includegraphics[height=8cm]{ptrbdc2.ps}}
1064: \caption{Evolution of price, traffic and revenue in the BDC regime, next to lowest class of service for $\alpha = -2.5, s = 2/3, \beta = 6$ and $a = 0.7$}
1065: \label{fig:ptrbdc2}
1066: \end{figure}
1067: The ratio of prices in two neighboring classes
1068: is interesting to study.
1069: The ratio between the prices of service in a class and the class just below,
1070: as a function of $c$, has the same behavior and the same values
1071: (for different $c$'s) for all classes.
1072: Let the classes of services are described by qualities:
1073: \mbox{$q_{0} < q_{1} < \ldots < q_{i} < \ldots$} and appearing
1074: for the values of $c$:
1075: \mbox{$c_{0} > c_{1} > \ldots > c_{i} > \ldots$}.
1076: Let \mbox{$p_{0}(c) < p_{1}(c) < \ldots < p_{i}(c) < \ldots$} be the prices
1077: of the corresponding services.
1078: The behavior of \mbox{$p_{1}(c) / p_{0}(c)$} for $c$
1079: between $c_{0}$ and $c_{1}$
1080: is the same as the behavior of \mbox{$p_{2}(c) / p_{1}(c)$} for $c$ between
1081: $c_{1}$ and $c_{2}$.
1082: The price of a newly created class (of high quality)
1083: drops (with $c$) more rapidly than that of an older class of lower quality
1084: and therefore the ratio \mbox{$p_{i+1}(c)/p_{i}(c)$},
1085: larger than $1$, decreases with $c$.
1086: Under the assumptions above, it approaches
1087: \[
1088: \left( \frac{s}{(2+\alpha)\: s - \alpha - 1} \right)^
1089: {\frac{(2+\alpha)\: s - \alpha - 1}{\alpha + 1}}
1090: \]
1091: when $c$ approaches zero.
1092: For \mbox{$\alpha = -2$} and \mbox{$s = 1/2$}, this gives a value of $2$,
1093: the ratio used by the Paris Metro system when it offered two classes
1094: of service, and quite typical of other transportation systems.
1095: \subsection{The UDC sub-regime}
1096: \label{subsec:UDCregime}
1097: We may now understand what happens in the insensitive case:
1098: the left hand side
1099: of~(\ref{eq:eq}) grows without bound. In this case, for any $c$,
1100: an infinite number of qualities: \mbox{$q_{0}, q_{1}, \ldots , q_{i} , \ldots$}
1101: of service is offered. The qualities offered are constant, they do not
1102: depend on $c$. Figures~\ref{fig:ptrudc1} and~\ref{fig:ptrudc2}
1103: show the first two classes for a generic case.
1104: \begin{figure}[htbp]
1105: \centering
1106: \ifthenelse{\boolean{color}}
1107: {\includegraphics[height=8cm]{ptrudc1col.ps}}
1108: {\includegraphics[height=8cm]{ptrudc1.ps}}
1109: \caption{Evolution of price, traffic and revenue in the UDC regime, lowest class of service for $\alpha = -2.5, s = 2/3, \beta = 2$ and $a = 1$}
1110: \label{fig:ptrudc1}
1111: \end{figure}
1112: \begin{figure}[htbp]
1113: \centering
1114: \ifthenelse{\boolean{color}}
1115: {\includegraphics[height=8cm]{ptrudc2col.ps}}
1116: {\includegraphics[height=8cm]{ptrudc2.ps}}
1117: \caption{Evolution of price, traffic and revenue in the UDC regime, next to lowest class of service for $\alpha = -2.5, s = 2/3, \beta = 2$ and $a = 1$}
1118: \label{fig:ptrudc2}
1119: \end{figure}
1120: \subsection{The general picture}
1121: \label{subsec:genpic}
1122: \begin{law}[Second Law]
1123: Under a decrease in the technological constant $c$, i.e., under technological
1124: progress and a drop in the price of equipment:
1125: \begin{itemize}
1126: \item In the UC regime, at any time, i.e., for any value of $c$,
1127: a unique class serves requests for arbitrary quality.
1128: \item In the DC regime,
1129: \begin{itemize}
1130: \item in the BDC sub-regime, no service is provided at first,
1131: then a service of finite quality is provided,
1132: later on another service of higher quality will be
1133: proposed, and so on. For any quality $q$ a service of quality $q$ will
1134: be proposed at some time (and on), but at any time there are qualities
1135: of services that are not provided.
1136: \item in the UDC sub-regime, at any time, an infinite number of classes of service
1137: are proposed and requests for arbitrary quality are taken care of.
1138: \end{itemize}
1139: \end{itemize}
1140: \end{law}
1141: This Second Law contradicts the folk wisdom that says that the Fat Dumb Pipe
1142: (FDP) model
1143: for the Internet will collapse when prices become low enough to enable
1144: very high quality services. It says that if the FDP model is the
1145: one that prevails when $c$ is high, it will continue to prevail after
1146: a decrease in $c$.
1147: A drop in the price of equipment cannot cause the breakdown of a system
1148: based on a universal class of service.
1149:
1150: \section{Further work and conclusions}
1151: \label{sec:conclusion}
1152: % Second order approximation of the demand curve
1153: % Real Data
1154: The most intriguing challenges are a theoretical one and an experimental one.
1155: The first one concerns the study of the case in which the Decoupling of
1156: Equation~\ref{eq:decoup}
1157: does not hold, or holds only approximately.
1158: The second one is the study of real systems, in particular the Internet,
1159: and the evaluation of the parameters of interest for them.
1160:
1161: The main conclusion of this work is that the parameters that determine
1162: the type of the prevailing market are the size of the economies of scale
1163: and the distribution of demand over the range of qualities.
1164: The sensitivity of demand to price and the changes in the price of
1165: equipment bear a lesser influence.
1166:
1167: This paper made two severe assumptions.
1168: It assumes perfect competition, and it assumes that economies of scale
1169: do not aggregate over different qualities of service. Further work should
1170: relax those assumptions.
1171: \bibliographystyle{plain}
1172: \bibliography{../my}
1173:
1174: \end{document}
1175: \bye
1176: