CA2160393C

CA2160393C - Method for integrated congestion control in networks

Info

Publication number: CA2160393C
Application number: CA002160393A
Authority: CA
Inventors: Anwar Elwalid
Original assignee: AT&T Corp
Current assignee: AT&T Corp
Priority date: 1994-12-30
Filing date: 1995-10-12
Publication date: 1999-05-04
Anticipated expiration: 2015-10-12
Also published as: JP2960004B2; EP0720414A2; CA2160393A1; EP0720414A3; US5646943A; JPH08251202A

Abstract

An integrated method for congestion control uses access regulator to control the admission of information from a communication device into a network according to a predetermined function characterized by a set of parameters. A node within the network sends a signal indicating a level of congestion in the the node to the access regulator, and at least one of the parameters is adjusted in response to the signal. The parameter may additionally be adjusted according to other parameterswhich which are selected according to stability and transient response criteria.

Description

METHOD FOR INTEGRATED CONGESTION CONT~OL IN NEIWORKS
Technical Field The invention relates to methods for controlling congestion in networ~s.
Background of the Invention Networks are a principal means of exch~nging or transferring information (e.g, data, voice, text, video, etc.) among cornmunication devices (i.e.
devices for inputting and/or outputting information such as coluL~uLe~ t~.rmin~
mllltim~ workstations, fax m~hines, printers, servers, telephones, videophones, etc.) connected to the network. The use of networks for the reliable, high speed10 transport of information between culllLuu~lication devices over wide areas in digital format is bringing major changes to network services and network arc'nitecture/infrastructure design. Many new services having diverse characteristics (e.g. different bit rates) are now feasible. One emerging example of such a network is Broadband ISDN (BISDN). BISDN is de~ign~d to support a variety of 15 applications such as interactive and distributed audio, video and data co"".,u~ications. ~he principal transfer mode for BISDN is called Asynchronous Transfer Mode (AFM). ATM is a high bandwidth, low-delay, packet-lil~e switching and multiplexing technique. See, S. E. Minzer, "Broadband ISDN and Asynchronous Transfer Mode," IEEE Comm. Mag., pp. 11-24, Sept. 1989.
A network typically compri~e~s switching nodes connected to each other, and to co"""ui~ tion devices, by links. Each link is chara~t~ri7e~ by a bandwidth or link capacity. I~ ~Lion input from the co"""~ tion devices to the network may be of any form but is often form~tte~ into fixed-length packets or cells. When information is to be ~x~h~nged between two co"""l",ication devices, a path 25 (comprising a set of nodes and links connecting the two col~ ullication devices) is established. This path is a so-called "virtual circuit" (VC) by which it is meant that a co"""ll"ication device simply specifies the ~estin~tion for the information, and the network delivers the information as though a dedicated circuit connected the twocommunication devices. Although a number of dirre~ t paths could be chosen 30 initially, once established, the path typically remains established until theinformation exchange is complete. Cells in transit between co,,,,,,ll,~ic~tion devices may t~lllL)oldlily be stored in buffers at nodes along the path of the virtual circuit pending sufficient available bandwidth on an outgoing link from the node.

An illlL~ol Lant consideration in network operation is traffic control, i e.
the process of controlling the arnount of inforrna~ion e~tering and traversing the network. Traffic control m~ch~nicmc are of three general types, each with dirrt~ t objectives. See, generally, William St~llingc, "Data and Co.l,~uter 5 Communications," 4th Ed., McMillan Pub. Co., New York, 1991. The first type oftraffic control, flow control, deals with the regulation of the rate of data tr~n.smicsion between two points. The second type of traffic control is deadlock avoidance in which various techniques are used to minimi7e conditions under which a node is unable to forward packets because no buffers are available in a subsequent node 10 along the path. Congestion control is the third type of traffic control. The objective of congestion control is to m~int~in the amount of information (e.g. the number of cells) within the network below a level at which pe,r~.l,uance (e.g. throughput delay, cell loss, etc.) becomes unacceptable. More particularly, the objective of congestion control is to limit the fullness of buffers at nodes in the network so as to avoid cell 15 loss or delay. Thus, congestion control may require overhead in that a portion of the capacity of the network is ~e~lir~t~l to calIying congestion information rather than to carrying information input from co""""";r,~tion devices.
Congestion control may be of the ''open-loop" type of control in which a requested VC is established within the network based on the requested VC's 20 declared parameters or characteristics (e.g. mean rate, peak rate and burst size), the net~,vork state (i.e. the network topology and current available network resources such as buffer space in the nodes and capacity in the links) and given any quality-of-service requirements (e.g. ~,u~l~ed bandwidth or m;-xi",l"" cell loss probability) made to users of devices in the network. The VC is then monitored in 25 real time, as for e~m~ by an access regulator, and is restricted to conform to its declared parameters. See, e.g. A.E. Eckberg, et al., "An Approach to ControllingCongestion in ATM Networks,'' International Jounzal on Digital and Analog Communication Systems, Vol. 3, pp. 199-209, 1990. Open loop control, however, does not perrnit a system to modify the declared parameters to react to (i.e. it is non-30 dynarnic with respect to) chzlnging network conditions (e.g. to take advantage of anincrease in the bandwidth available on a link at a node due to rennin~tion of other VCs routed through the node).
Congestion control may also be of the "feedback" type of control as used in data networks in which a node generates or controls a feedback signal which 35 is used to indicate a level of congestion in the vicinity of the node. The congestion level is indicative of the extent to which a cell arriving at the node may be delayed or lost within the node. Thus, a level of congestion may be reflected by the number of cells in transit through the node, the content level of a node bufirer, the throughput delay at the node, the probability or amount of cell loss at the node, etc. rhe feedback signal is received al a communication device, and then the cornmunication 5 device regulates the admission of new informalion into the network in response to the level of congestion indicated by Ihe feedback signal. One such technique is described in D. Mitra and J. B. Seerv, "Method for Adaptive Control of Windows and Rates in Networks," U.S. Patent 5,426,635 (issued June 20, 1995) and assigned to the assignee of tnis invention. See also, R. Jain and K. Rarnakrishnan, 10 "Congestion Avoidance in Computer Networks wilh Connectionless ~etwork Layer:Concepts, Goals, and Methodology," Proc. Computer IVerworking Symposium, April 1988. However, tradilional data networks typically do not perform functions to control adrnission of new VCs into the network. Hence, when many VCs are admitted, excessive tnroughput delavs may be encountered. Additionally, feedbac~15 congestion control methods may encounter stability problerns in cases where the feedback is delayed in reaching the communication device.
Sl~mm~ry of the Invention rhe inventive method recognizes that the open-loop and feedbacl~
congestion control techniques can advantageously be integrated to overcome the 20 non-dynamic nature of open-loop control and the delay and stability problems of feedback control . The method regulates, according to a predetermined function characterized by at least a first parameter, admission of information from a communications device to a network and adjusts the first parameter in response to a sianal received from a node in the network where the signal indicates a level of25 congestion at the node. In preferred embodiments, an access regulator controls the admission of information to the ne~vork according to the predetermined function. In an aspect of the invention, the predetermined function of the access regulator is characterized by a token rate parameter where the token rate parameter is set responsive to the signal. In another aspect of the invention, the feedback signal 30 indicates a difference between the level of congestion at the node and a target level, and the rate parameter is adjusted as a function of the feedback signal and other selected pararneters. ~hus, the inventive method solves shortcomings of prior methods in that the inventive method reacts to ch~n~ing conditions ~vithin the network, and proper choice of the selected paramelers insures stability of the 35 inventive method in controlling congestion in the network.

3 ~ ~
- 3a-In accordance with one aspect of the present invention there is provided a method of setting a value r for a token rate parameter of an access regulator, wherein said access regulator controls admission of information from a communication device to a network and wherein said network comprises a node, 5 the method comprising the steps of: receiving a signal at said access regulator from said node, said node comprising a buffer characterized by a buffer content level, said signal indicating the buffer content level of said node relative to a target level x, and setting said value r for said token rate parameter at a time (t+~) based on a difference of a first value based on the rate r(t) at time t and a second value 10 based on said signal indicating the relative buffer content, respectively, wherein said first value corresponds to the rate r at time t scaled by the complement ofdamping parameter ~ as scaled by time difference ~ and said second value corresponds to said relative buffer content scaled by gain parameter a and the time difference ~, wherein the damping and gain parameters ~ and a substantially satisfy 15 the expression sin ~ <
~ a~

where ~ tan cl) and I is a propagation delay between said node and said access regulator.
In accordance with another aspect of the present invention there is provided in a network comprising a node and a set of K information sources, 20 wherein the kth source in said set of K information sources is connectedto a respective kth access regulator, wherein each access regulator is connected to said node, and wherein the k'h access regulator is characterized by a rate, rk(t), at which information from the kth source is admitted to said network by said kth access regulator, a method comprising the steps of: receiving a signal at each access 25 regulator from said node, said node comprising a buffer characterized by a buffer content level, said signal indicating said buffer content level relative to a target value; and setting the rate rk(t) of said kth access regulator at a time (t+~) based on a difference of a first value based on the rate rk(t) at time t of said k'h access regulator and a second value based on said signal indicating the relative buffer30 content, respectively, wherein said first value corresponds to the current rate r(t) - 3b-scaled by the complement of damping parameter ~ as scaled by time difference ~
and said second value corresponds to said relative buffer content scaled by a gain parameter ak and the time difference ~, wherein the damping and gain parameters y and ak substantially satisfy the expression sin a) _ < ~y , ~ak~

~=1 5 where ~ = ~ tan c~ and I is a propagation delay between said node and said access regulator.

, ,~.

21 603~3 Brief Des~ tion of the Drawin~s FIG. 1 illustrates a network in which the inventive method may be practiced.
FIG. 2 illustrates the components of a leaky bucket access regulator.
FIG. 3 illustrates operation of an embodiment of the inventive method in a network.
FIG. 4 illustrates the steps in the inventive method.
FIG. ~ illustrates a set of regions in which the ratio of a damping constant to a gain parameter may lie.
FIG. 6 is a plot of a norm~li7e~ critical damping parameters as a function of a gain parameter.
FIG. 7 is a plot of a principal root versus a damping constant.
FIG. 8 is a plot of a relationship between delay, throughput and dynamic range based on a set of parameters.
FIG. 9 illustrates operation of the i~vellLLve method for multiple communication devices connected to a node.
De~ailed Description Introduchon ~he detailed description of the inventive method for congestion control ~0 is divided into three sections. The first section ~l~,sell~s an overview of the il~Ve~l~,iVe method for congestion control in a node in a network. This overview first illustrates operation of the inventive method at a node in a nelwol~ for a single co"""-l.,;c~tion device (or source) connected to the node where information from the source to the network is regulated according to a pre~t~rmin~d function ch~a Ie~ ;,,e~ by a set of 25 parameters. For illustrative purposes, the regulation of infomlation is by an access regulator. The overview then specifies how access regulator parameters characterizing the predet~rrninefl function may be advantageously adjusted as a func~ion of a feedback signal from the node. In particular the overview also describes how a rate parameter char~t~ri7ing an access regulator may be adjusted30 where the feedback signal in~ res the difference between a buffer content level of the node and a target value. In a feature of the inventive method, the rate parameter is also adjusted according to a darnping constant and gain parameter, and the selection of the gain constant and damping parameter is discussed. Finally, the overview illustrates an application of the inventive method to a situation in which 35 multiple sources are connected to a node. The second section of the detailed 2~ ~0393 description provides a more rigorous presentation of the inventive method. The final section provides a surnmary and conclusion.
Overview of a Methodfor Inte~,rated Con~estion Control FIG. 1 illustrates a networl~ in which the inventive integrated congestion control method rnay be practiced. Network 110 comprises access ~wiLchi~lg nodes 130 - i and links 140 - k. Network 110 also comprises means 120 -j for regulating, according to a pre~et~rmined function characterized by at least one parameter, ~1mi~sion of information from communic~tion device 105 -j to network 110. In preferred embo~im~nt~, means 120-j is an access regulator. For illustrative 10 purposes, access regulator 120--j will be considered to be a leaky bucket regulator (LBR) and hereinafter will be referred to as such. However, that other types of access regulators known in the art, e.g. buffered leaky bucket regulators or c~c~de~
LBRs, may be used. See, e.g. A.I. Elwalid and D. Mitra, "Analysis and Design of Rate-Based Congestion control of High Speed N~lwo~ Stochastic Fluid Models, 15 Access Regulation," Q~erleing Systems, T~eory and Applic~tzon, Vol. 9, pp. 29-64, 1991. Retllrning to FIG. 1, communication device 105-j generates information foruse by, or receives information from, other comml-nic"tion devices in the network.
By information it is meant data, text, voice, video, etc. Comm-lnication device 105-j may include such apparatus (e.g. analog-to-digital converters) so as to render 20 the information suitable for tr~n~mi~ion on network 110.
The operation of leaky bucket regulator 120-j is shown conceptually in FIG. 2. Leaky bucket regulator 120-j comprises switch 222-j and buffer 224-j.
LBR 120-j accepts information from co,,,,,,ll.,ir~tion device 105-j. The information from co"~l"ll~-ir~tion device 105-j is illustratively form~rte~, into 25 packets or cells. Switch 222-j is closed, and therefore data is output from LBR
120-j only when a token is available in buffer 224-j. Buffer 224-j is of size BTTokens are supplied to the buffer at rate rj, and a token is used, and thus is no longer available in the buffer, each time a specified amount of information (e.g. a cell) is output from the LBR. Information is output from LBR 120-j at a m~ximllm rate of 30 Pj. Thus, LBR 1'70-j regulates the adrnission of inforrnation according to a function of a set of three parameters: the token rate rj which bounds the long-term average rate of inforrnation output from LBR 120-j, the token buffer size BTj which bounds the burst size of the inforrnation, and a peak rate Pj which bounds the rate at which information may leave LBR 120-j.

In the iLlu~lldLive LBR 120-j of FIG. 2, if co~ ul~ication device 105 -j inputs ce~ls to I BR 120-j at a rate so great that the buffer underflows, cells will be lost. However, the LBR may be of the type in which, when no tokens are available, cells are "rnarked" as low priority cells; the marked cells are then routed through the network on a VC but are first to be dropped if congestion is encountered at nodes along the VC. Thus, regardless of the type of regulator used, communication device 105-j must contract with the network when establishing the VC for access regulator parameters sufficient to ensure a maximum tolerable cellloss.
Re~nningtoFIG. 1,inorderforcou"uunicationdevice 105-1 to transmit information to cornmunication device 105-2, a virtual circuit (VC) needs to be established between them. l'he VC is a path comprising a set of nodes and set of links between the L BRs associated with communication devices 105-1 and 105-2.
For example in FIG. 1, the VC betwee~n co~ ullication device 105-1 and 105-2 15 could comprise nodes 130-1, 130-2, 103-3 and links 140-1, 140-3, 140-5 and 140-7.
Each node 130-i has associated with it at least one buffer of size BNi, and each link has associated with it a bandwidth capacity Ck.
FIG. 3 illu~ tes operation of an embodiment of the inventive method in a network of the type in FIG. 1. A VC comprising links 320- 1 and 320-2 is routed 20 from co"""ll";c~tion device 301-1 through node 310 (ie. co-"",l~,-ic~tion device 301-1 acts as a "source" of inforrnation). Link 320-1 is char~e i7ed by a propagation delay ~ l between source 301- 1 and node 310. Node 310 compri.~es buffer (or queue) 315 connecte~l to outgoing Link 320-2. Source 301-1 is monitored and regulated by ~$coci~te~1 access regulator 302-1 which is illustratively shown as 25 the leaky bucket regulator of FIG. 2, i.e. access regulator 302-1 comprices switch 322-1 and token buffer 32~1. Source 301-1 is capable of tr~n~mitting large bursts of information (e.g. as in file transfer applications).
Node 310 returns to access regulator 302- 1 (as for example via a bidirectional channel in the VC) a fee~ik~ signal co"r~it-il-g information on the 30 congestion status of the node buffer. The feedback signal reaches access regulator 302- 1 after a propagation delay of ~c ~ . The feedbacl~ information on the buffer status may be explicitly obtained using forward and backward congestion notific~tions or implicitly from time response measurements as is well known in the art. See, e.g.
K.W. Fendick et al., "Analysis of Rate-Based Feedback Control Strategy for Long 35 Haul Dala Transport," Perfonnance Evaluanon, Vol. 16, pp. 67-94, 1992. Accessregulator 302- 1 has an associated set of parameters characterizing the flow of 2~ 60393 information from source 301-1 (e.g. the source's mean rate, peak rate and variability/bllr~rin~sc). Each source's pa~ are negotiated between the source and the network and are the basis of the establishing of VCs requested by the sources to the network.
S One parameter of leaky bucl~et regulator 302-1 is the token rate r I
which controls the average rate (throughput) of info~nation output from LBR 302-1.
In a preferred embodiment, the inventive method controls the token rate r according to the rule:
dt r l (t) =-a[x~t -~2)--x~ -yrl (t) ( 1 ) 10 which, in the discrete time context where d rl (t+~)-rl (t) d rl(t)-- ~

for small ~, where A is a period of ~ rt~tion~ can be written as rl (t+~)=rl (t)[1-~l-a[x(t-~2)-~ (2) x is a target value for the content level of buffer 315, x is the content level of buffer 15 315, a is a gain parameter and y is a damping consl~l. The above equations indicate that the rate of information output from an access regulator can be adapted to the congestion level at a node by regulating a parameter (e.g. the token rate) of the access regulator, where the congestion level at the node is measured relative to a target value for the content of the buffer node. Thus, an a~loL,liate choice of x, a, 20 and ~ will result in an increase (decrease) in r(t) when congestion is low (high). In short, the pa~d~t;~el~ of the access regulator are adjusted so as to I~ t~ an acceptable queuing delay at buffer node 315. The rate rl can advantageously be limited to a range in which the ",i,~i"""., value of rl is the value initially contracted for when the VC was established. The maximum value of r I is advantageously ''S selected based on desired dynamic range and transient response characteristics as shown below.
FIG. 4 illustrates the steps in the inventive method. For convenience, as explained in the rigorous presentation of the inventive method in the next section, a new gain parameter A, A = ~2 a, and a new damping constant r r = ~~r, will be 30 introduced, where ~ + ~ 2 ~ Assuming a VC has been established, values of A and r are selected in step 420. Once selected, these values in the preferred embodiment typicaLly remain constant for as long as the VC is established. The values of A and r 2 1 60}93 are selected according to several conditions as illustrated in FIG. 5. First, the ratio of A/I is advantageously selected so that for a given value of A, r2rcn, where rcntdefines the boundary between stable and unstable operation of the method as shown by line 510 in FIG. S and by line 610 in FIG. 6~ As explained below, stable 5 operation means asymptotic, non-oscillatory steady state behavior where values of r equal to, or in excess of, a critical value rcnt provide such stable behavior. The ratio of A/r is further advantageously selected according to a desired transient response.
In particular, for a given value of A, an optimal value of r termed rOpr is given by line 520 in FIG. 5. ropt is the value of r for which the transient behavior is "best" in 10 the sense that the asymptotic approach to steady state is both the fastest and oscillation-free.
As explained in the next section, the value of r actually selected need not be rOpt. In particular, the ratio of A/r may be adjusted so as to trade dyn~mic range (i.e. the range over which the rate r(t) tracks available link capacity at the 15 node as ~etect~Pf~ by decreases in buffer content) for transient response. In FIG. 5, for exarnple, a decrease in the value of r such that rcr~t<r<rOpt, (i.e. a value of r in the region below line 520 and above or equal to line ~10) will result in an increased dynamic range but a in a slower transient response. Selection of r > r opt results in both a decrease in dynamic range and a slower transient response.
In step 430 a target value x is selected for the buffer level of a node through which the established VC is routed. A large x collc~ollds to a large delay.
The value of x is selected according to the desired delay through the node where the delay is advantageously ~et~ ned in terms of: the difference between the contents y of buffer 315 and the target value; the capacity C of an outgoing link conn~cted to ~5 the node; and the fraction F of the capacity guaranteed to the established VC which is related to the token rate initially contracted for. In short, delay = C(y -x~ F. As shown, below, the fraction of the link capacity guaranteed to a VC is a function of the gain parameter. ~ItPm~tively, the target value may also be dyn~mic~lly adjusted by an intelligent node in order to illlpl~ve the steady state p~,~rol,llance and dynamic 30 range.
Renlrning to FIG. 4, in step 440 a feedback signal, indicating a level of congestion at the node, is sent to an access regulator from the node and using the equations above, one or more access regulator parameters, e.g. r(t), may be adjusted within the access regulator in step 450. This process then advantageously continues 35 for as long as the VC is established.

The inventive method extends to the case where K access regulators are connecte~ to a node (i.e. there are K VCs routed through the node). Let f a I, a ~, - -- . a K~ be a set of gain parameters where a ~ is the gain parameter of the i ~h connection. With a = ~, a i and with the damping pararneter y common to all 5 connections, A and r and x can be chosen as above. T'nen d r l ( t) = - a l [x( t ~ yr l ( t) d r2(t)=-a~ [X(t-~2)--~l-yr2(t) dtrK(t)=--aK[x(t--~ x~--yrK(t) 10 The fraction of the link capacity guaranteed to the i 'h VC is--. Note that the fraclion of the bandwidth allocated to the i'~ VC may exceed--if there is unused ~ink capacity. A more rigorous detailed presentation of the above method is presented in the ne~t section.
A Method For Inte~rated Congestion Control In this section a rigorous explanation of the inventive method is presPnte-1 In particular, the criteria for asymptotic stability of the method in terms of the gain pa~ter and damping constant are pres~nrt-~l and the trade-off be~veen steady state behavior and transient response is detailed. The method is also extended to the case of multiple access regulators connected to a node.
~0 171e Model First, consider a single connection between a source controlled by an access regulator and a node, see FIG. 3. Results for the single connection case generalize to the multiple connection case which is treated below. The source rate S(t) iS controlled by the token rate r(t) and its maximum burst length is controlled '5 by the token buffer size BT The regulated rate Q(t) depends on the token buffer content y(t). The buffer content of the node is x(t) and the available tr~n~mission c~pacity is u. Associated with the distant buffer is a target value x- above which the node is considered congested. The source rate is s ( t), the content of the token buffer 21 6~393 is y(t) and the output of the regulator is ~(t). The propagation delay from the source to the node is ~ 1 and the propagation delay from the node to the source is ~ 2 . The objective is to advantageously and dynamically adapt r(t) to ~L while keeping x(t) at an acceptable level based on feedback information on the the status of the buffer S content which is received by the source. The available channel rate or capacity, !;L, is typically a function of cross traffic from other connections and therefore is unknown.
As shown below, where multiple connections share the node, u varies as connections are set up and tennin~te-i Next a class of methods is considered where the source rate varies in 10 proportion to the difference between the buffer content and the target value x. The data traffic is modeled as ~uid flow. The applicability of fluid models to congestion control in ATM has been previously demonstrated in a stochastic setting and in adete~mini~tjc setting. The constancy and small size of the ATM cell makes its tr~n~mis.cion time small co~ ;d to the total transfer tirne and justifies the fluid 15 approximation. See, A.I. Elwalid and D. l~itra, Fluid Models for the Analysis and Desgin of Statistical Multiplexing With Loss Priorities on Multiple Classes of Bursty Traffic," Proc. IEEE INFOCOM '92, pp. 415~25, 1992; K.W. Fendick, et al.,supra.
The following set of coupled delay-di~r~,iential equations describe the 20 system evolution, d ~(t~ if x(t)>O
dtX(t) = ~(~(t-~ )+ if x(t)=0 (1) where, r(t) if y(t)--0 ' (2) d r(t) =-a[x(t-~2)--~-vr(t), (3) 25 and d r(t)-s(t) if y(t)>0 dt (r(t)-s(t))t if y(t)=0 ( ) where the positive qu~ntitieS a, Ay and x are the gain parameter, the damping constant and the buffer target, respectively.

2~ 60393 It has been shown that the throughput of the access regulator is an increasing func~ion of both the token buffer size B T and the token rate. Increasing B T~ however, has the added effect of increasing the burstin~-s.s of the regulated rate by increasing the period during which the regulated rate is equal the source peak 5 rate. During this period the surge of rraffic may flood network nodes. I~he throughput of the re~ulator, therefore, may be advantageously increased primarily through r( t) to achieve smooth increase in the traffic rate. The access regulator may have a data buffer to ~uldlily store data when the token buffer is empty and thesource rate is at its peak. It has been shown that the regulator throughput depends on 10 the token buffer size and data buffer size only through their sum. The token buffer can thus be made small at the expense of having larger data buffer. See, Elwalid and Mitra, supra.
Henceforth, B T can advantageously be considered to be small and fixed, and consider the adaptation of r( t). The token buffer comes into play only when the 15 source turns from off to on. In this case the output of the regulator will be equal to the source peak rate until the token buffer empties, and at which it becomes equal to r(t). I'he token buffer has the effect of adding a small burst of peak rate at the beginning of each tr~n~miscion period. Ihe tr~ncmi~.~ion periods are typically long in applications requiring large data transfers and the initial peak rate duration will be 20 negligible. I'his burst period is advantageously ignored (its effect may be approximated by setting the initial rate of each tr~n~mis~ion period to the source peak rate.) ~ItP~n~tively, the source may be ~c~nmP~ to be always on (infinite data source) to isolate the issue of adaptation m~ch~ni~m form other considerations.
According to the above observation, let ~( t) = r(t) and thereby the 25 equations are d ~(t~ if x(t)>O
dt ~(d)(t-~l )--U)+ if X(t)=O

d ~(t)=-a[X(t-~2)--X-l--~yq)(t) (6) Stability Analysis Letting x(t)=x(t)-x and (I)=(i)(t)-U and combining (5) and (6), yields 2 ~(t)+~d ~(t)+a~(t-~)=0 (7) where ~ +~ is the round-trip propagation delay. Henceforth, let the unit of 2 t 60393 time be ~ and use the new time scale t' = t/~. According to this norrn~li7~tion,equation (7) becomes dt2 '~( ) dt ( ) ( ) (8) where r=rrandA =~~a. The charactenstic equation associated with (8) is s (rS +52) eS +A =O (9) The transcendental Equation (9) has infinite number of roots and the location ofwhich in the complex plane determines the behavior of the equilibrium point of the dynamical system in response to small perturbation. In addition, any solution to (9) can be represented by an a~yll.yLoLic series, 10 - ' s~' (10) n = I
Definiffon: The system is asymptotically stable if:
1- It is stable in the sense of Lyapunov.
2- Every initial solution starting near (~,~ ,x,O ) co~ rOes to (~OO ,xOO ) as t~, where ( ~ O~ ,xOO ) is the equilibrium point.
For asymptotic stability all roots of (9) must have negative real parts.
To find conditions on system par~m~tçr~ which guarantee a~y~ Lic stability, certain ~ln~l~m~nt~l results on the zeros of exponential polynomials due to Pontryagen can be used. See, e.g., S.J. Bhatt and C.S. Hsu, "Stability Criteria for Second-Order Dyn~mic~l Systems With Tirne-Lag," Trans. ASME, Vol. E-33, pp.
20 1 13-1 18, 1966.
Consider the ~eneral exponential polynomial n m P(s,e ) = ~ ~ bkl sk leS~l =0 ( l 1 ) k=O l=O
Let the proper degrees of P(s,es ) with respect to s and eS be lV and M ~ el;~ively.
The terrn b ,VM5N f eS ~M iS called the principal term of P(s,es ). A necessary ~5 condition for an exponential polynomial to have all its roots in the left half plane is that it has a principal term. If s = ic~, equation ( 11) can be written as P(i~,e'~)=F(~)+iG(~) (12) Theorem 1. ~ecessary and sufficient conditions for an exponential polynomial with 21 603q3 principal term to have stable roots are ~i) G(~) [or F( ~)] has exactly 4kM +N real zeros in the interval -2k~+~5~3<2k~c+~ starting with sufficiently large k, and ~ being some a~plo~liate constant.
5 (ii) For each zero of G(~) [or F(~))], denoted by ~i, F(~i~ d~ G(c~i) >O ( 13) or, -G(~i) d~F(~ ~ (14) From (9), F( ~) = - ~( ~ cos~+ r sin ~) +A ( 15) G(~) = _~2 sin ~+r~3 cos ~ (16) For stability the values of r and A which will insure that G(~) has the a~ploL,liate number of real zeros must be found. Since the princir~l term in (9) is (52 eS )~ G(c~) must have exactly 6 real zeros in the interval -2k?~+~C2k7~+~. It is obvious that (d = O is a root of ( 16). To find the rest of the roots, the solutions of d = rcot~3 are 15 e~c~mine~i If k = 1 and ~=1~/2, then there are 5 roots in (-311/2,51IJ2). Thus, there are a total of 6 zeros of G(c~) as required. From condition (ii), the following inequality must apply to all roots of (16), [(r-~2)cos~-(2+r)a)sin~][-~32cos~-r~sinc~+A]>0 (17) For ~=0, (21) implies that rA ~0, which is col~c;.~ilte~lt with the assumption thatA
20 and r are posiLive. To apply condition (17) to other roots, it is written as S~ "4 -r2 ~ -r~d2)( _~4 -~d' r2 +rA ~ ) >o ( 18) which implies that -~d4-C32r2+rA ~ <o (19) sm ~3 Since ~ is a zero of (16), (19) becomes A n~ <r (20) G~
Note that in (20) d~ [0,~] gives the tightest bound on r for a Oiven A. Equation(20) is satisfied for all zeros fc~i~ of G(~) if and only if it is satisfied for ~ [0,7~].
Hence, 5 Theorem ~. The delay-diLrele,lLial equation (7) is asymptotically stable if and only if sin~
A <r (21) ~ 1 where ~d I is the unique solution of ~=~tan~. (22) 10 in the interval (0,~). FIG. 6 gives a plot of the critical damping rcn~ as a function of A. where rcnt is the i"ri,,,l,,,, of values of r satisfying the stability conditions for a glven A.
Przncipal Root and Rate of Sta~ilization In this section the rate at which a stable system approaches steady state 15 is detorminel1 In this connection consider the principal root s *, which is the root of (9) having the largest real part. It then follows from (10) that ~(t' ) -Cei ' (23 ) i.e., the ratio of the left hand and the right hand sides asymptotically approaches unity. Also from (10) ~0 ¦ ¦~(t~ ceRe~S )' t~>o (24) where c is a suitably chosen constant taking into account initial conditions.
Let s ~ = ~+i ,B, it then follows that ~ can be taken as the time constant of the dynamic method, i.e., the time it takes for a small per~urbation around the steady-state value to decrease by a factor of e~ I . Prior work which found the ~5 principal root of a system of delay-~ r~le.l~ial equations and the principal root for ~rst-order delay-differential equations in a different context are: D. l~/Iitra and A.
Weiss, "Analysis of Dealy Differential Equations Arising in Commllnic~rion Network Synchronization," Proc. IEEE Int. Symp. Circuits Syst. pp. 839-843, April 1980, D. Mitra "Networl~ Syncl~ aLion: Analysis of Hybrid master-Slave and ~1 603~3 Mutual synchronization," IEEE Trans. Comm., Vol. COM-28, No. 8, pp. 1245-1259, 1980, and F. Brauer, "Decay Rates for Solutions of a CIass of DiLr~l~,ntial-Difference Equations," 5~4M J. rrulth. Anal., Vol. 10, No. 4, pp. 783-788, 1979.
First derermin~ ~, the real part of s ', and then de~e~ "~ine the im~gin:~ry 5 part ~. The change of variable s = s ' - 6 transforrns the characteristic equation (9) to [s' +s' (r- 26~ +(62 -6r)] eS +Ae~ =() (25) If 6 is chosen to be the ~U~ lllUlll of positive real numbers for which the transformed characteristic equation (25) has all roots in the left half plane, then ~=- 6. The conditions for the roots of (25) to be in the left half plane can be 10 derived using Theorem 1. rhese conditions are given as a set of separate inequalities:
6(~-r)<0 (26) (26-r)<6(6-r) (27) 6(r-6) <Ae 5 (28) --Ae ~ c (r - 2 6) (29 ) where ~ I is the unique solution of tan~ (~2 +6(r_6))(r-26) =~

in the interval (0,~). Now choose 6 to be the ~ul~r~ ul-l of positive real numbers satisfying (26)-(29). According to (29) alone, 6= r/2. Next consider inequality 20 (2'7) which can be l~wlilLe.~ as 62-6(r+2)+r>0 (31) Therefore, (31) is satisfied if 6< 6 l or 6> 62 where 62,l =6=(r/2+ I)_~l(r/2)I + 1 (32) Note that 6 l < r/2 < 62. Hence (27) restricts 6 to be in the range (0, 6 l ). Now 25 consider condition (28) written as 6(r-6) e~~ <A (33 ) The left hand side of (33) has a m~ximllm at 6 l and is monotonically increasing 2~ 60393 with 6 in the range (0.~ 1). Hence, if (33) is not satisfied for 6= 6 l then thesolution of (33) in 6, where the inequaiity is replaced witn equality, is the ~uylClllull ~
of ~ satisfying conditions (26)-(28); othenvise (33) (or (28)) is irrelevant (i.e., places no restrictions on 6.) Now consider condition (29). To investigate the behavior of ~ 1 as a function of ~, denote the left hand side of (30) byf(ld,6), ie.
tan~ (~2 +6(r_6))_(r-26)=o Observe that limf(~3 ~)=-(62-~s(r+2)+r). (35) ~ ,o 10 The right hand side of (35) is the negative of the left hand side of (30). Also, lim f(~,~) = ~ (36) ~,~ /2 iIn f(~3.6) = --~ (37) ~7~' f2 limf(~.~) = 0 (38) - ~D~rc Since af(~.~)/a~,o. it follows that there is a solution of (30) in (0,~/2), denoted 15 by ~1~ if lim",~0f(~,~)<0 (or 6<~l, where 6l is defined in (32)). Implicit diLre~ tiation of (34) gives d ~ l /d 6 < 0. Thus, ~3 l is monotonically decreasing func~ion of ~s.
Now consider equation (29) lewli~ below for ease of reference, sin~l Ae~ <r--26 (39) ~1 20 Note that increasing 6 in the range (~~51 ) increases the left hand side of (39) and decreases its right hand side. Hence, (39) could be relevant (imposes a restriction on 6). Recall that the largest 6 for which a solution ~ (0,~) of (30) exists is 6 1, giving ~ I = Ot. rhus, (39) is irrelevant if Ae ~1 < (r- 26 l )- If Ae ~' 2 (r - 26 l )7 the supremum of 6 satisfying (39~ is in (0,6 1) and can be obtained by solving (39) 25 with equality instead of inequality. Observe further that r-~6l =6l (r-~ ). (40) rherefore, if (39) is relevant then condition (28) is irrelevant and vice versa; the 2 l 603q3 relevant of the two ,gives the desired ~, where - 6 is the real part of the principal root s Now find ~, the im~oin~T y part of s * . If s = - ~+ i,~ is substituted in the characteristic equation (9) and the real and im~,o,in~ry parts equated, then ~(~--r)+Ae~cos~B--~2 = O (41) ~(r-2~)-Ae~sin~ = 0 (42) If cs~ (0,~l ) is obtained from (32) then upon substitution of (28) in (41) ~=0 is obtained. If ~ is obtained from (39) then upon substitution of (39) in (42) ,B = G~ I ~ (0,~/2) is obtained.
10 Theorem 3. If for a stable system the condition ~ 1 (r- 6 l ) >Ae ~' is satisfied, where ~ 1 is given by (32), then ~, the negative of the real part of the principal root, is the unique solution of ~s(r-~)=Ae5, (43) in ( 0, ~ l ), and ,13, the im~,~in~ry part of the principal root, is zero.
If the condition is not satisfied, then ~ is the unique solution in (0, 6 1 ) of sin~ l Ae~ =r-2~, (44) Cd where (d I is the unique solution of t~no [~2 +~(r_6)] -(r_2~) =0, (45 20 in (0,~/2), and ~
Dependence of the Principal Root on the Darnping Coe~cient In this section the ~p~n~ e of the principal root on the damping coefficient, r is ex~min~d when all other control parameters are held constant. The value of r which gives O~uulll transient behavior can then be ~. ",;,-~ From 25 Theorem 3, if ~ satisfies (43), and hence ~=0, then implicit differentiation of (43) gives d -s/d r < 0. If ~ satisfies (44) then d ~;5/d r > O and d ,B/d r < 0. Therefore, Theorem 4.

2! 60393 if r<rOpt then dr >~ and dF~ <~

if r > rOpt then dr < ~ and ,13 =0 where rOpt is the solution to ~l (r-~l )=Ae~
5 and ~ 1 is defined in (32). The largest possible ~, denoted by ~*, is obtained when r = r opt and is given by ~ =(rOpt/2+ ~ (rOpt/2)2 + 1 Theorem 4 specifies the ~mping coefficient, rOpt, which gives the optimal transient behavior of the illv~uLiv~ method in the sense that the asymptotic 10 approach to steady state is fastest and oscillation-free.
Figure / depicts the princip~l root as a function of r.
Steady-State Response and Dynamic Range of Adaptation In this section the steady-state behavior of the mve~tlve method is studied. Two basic measures which ~es~ribe the steady-state perf ~rm~nre are the15 throughput and delay. The steady-state throughput is denoted by ~O, and the steady-state delay is inferred form xOO, the steady-state value of the buffer content. An oltant feature of the iuv~u~iv~; method is its dynamic range, which is Ihe rangeover which the input rate tracl~s the available channel capacity at the node. It will be shown that the three 4,,~ s: throughput, delay and dynamic range are 20 interdepen~l~nt Consider the iuverllive method with buffer floor, and assume here that the method with no buffer ~oor is made a~ylu~olically stable with an a~ylu~liate selection of the gain and damping par~m~rs For asymptotic stability,(dx(t)/dt) ~0 and (d/b(t)/dt) ~0. Hence, u=0 if x~,O>O (46) 25[~ U]+=0 if ~,,O=0 (47) -a(xOO-~)=~ (48) First consider solutions with xOO = 0. From (47~ and (48) ax (49) SO (50) Hence using (46) yields Ax/I'S ju. Thus one set of solutions to (46)-(48) is:

Ar Sll; Xoo=O; ~ =Ar (51) S Now consider solutions with xOO > O. Similarly, another set of solutions is obtained:

u ' r ; x = x-- A ; ~ oo = u.

Hence, the following result Proposition 1. The only steady-state solutions are fXoO=x-rAU; ~oo=U~ if,u<Ar (52) ~xOO =o; ~00 = Ar ~ if ~L2 Ar~ (53) Figure 8 illustrates the dependence of the steady-state solution on the control par~m~ters. First note that the optimal pelro"llaace in the sense of ma~cimi7ing throughput and minimi~in~ buffer content is obtained if ,u is equal to ~ . Thus, A~/r can be viewed as the upper limit of the dynamic range of the 15 method, i.e. the m~ximnm value that ~w can attain. Thus, if A~/I' < ,u, the case of insufficient dynamic range, ~,,0 is limited to Ax/r and x~O =0. If, on the other hand, A~/r > u, the case of excessive dynamic range, ~ and x~O > O. I'he op~ lu~ll performance is achieved when Ax-/r = ,u, giving ~ ~ = ,u and xOO = O. When there are multiple connections sharing the node, as is the case in practice, the residual capacity 20 available to a given connection ~L is ch~nging with time as connections are set up and tP~min~t~ If a large enough dynamic range is sought such that ~ then A~/r must be as large as u can get. The increase in A~/r will be achieved primarily by incre~sing x, since the ratio A/r is limited by the stability conditions. From Proposition 1 this implies large x~ . Therefore, there is a tradeoffbetween dynamic 25 range and steady-state behavior.
A~ultiple Connecrions:

Now extend the analysis of the hlvc:llLivt; method to the case where multiple co".."l,.-ic~tion devices share a node as shown in FIG. 9. The partition of the available capacit,v ,u in link 915 among commllnir~tion devices 901 -k connected to access regulators 905 -k wi31 be accomplished through a set of gain par~m~t~.5 fa I ,a2, . . ., aK~, where ai is the gain parameter of the ith co~ u~ication device, and K is the number of cc.,.,..,.l.-ic~.tion devices. The damping parameter ~ iscommon to all communir~tion devices. The node buffer in node 910 is shared by all commllnication devices and the service rlicr.il)lin~ iS ,.csllm~-3 FIFO. The following equations describe the evolution of the method d ~ i~i(t-~l )-U if x(t)>o dt (~;~i(t~ U)+ if x(t)=O (54) dt ~, (t) = -a l [X(t -~2) -~ -~ l (t) (55) d ~2(t) = -a2[x(t-~2)-x~-~y¢2(t) (56) t (57) (58) d ~(t)=-a.y[x(t-~2)--~ K(t) (59) The stability ct~n-lition of the inv. ~ive method is i<l~n*c~l to those obtained for the single co~ r~*on device case with (a = ~, i a i). There is a co~non rate of adaptation for all co,..""lllic~tion devices which is obtained from the pnncipal root of the char~t~ri~tic equation associated with the global system.
The steady-state value of each col"",n~ication device will depend on its gain pararneter. In partic,ular fx =x_ rU; ~ if !l~ r fx~. =o; ~ = r ~ if ,u>+ Ax (60) whereA~ =ai~2, r=rrandA = ~Ai.

21~393 conciusion This disclosure describes methods for integrated congestion control in networks. While the above teaching of the present invention has been in terms ofcontrolLing a pa~ ter of an access regulator, those skilled in the art will recogr~ize 5 the applicability of this teaching to other specific contexts. For e~r~mple, the damping constant and gain pararneters may be periodically updated and the accessregulator pararneters adjusted as a function of the updates.
The methods disclosed herein have been described without reference to specific ha~dw~e or soft~tvare. Tn~tto~(l the methods have been described in such a 10 manner that those skilled in the art can readily adopt such hal.lwale and software as may be available or preferable for a particular application.

Claims

Claims:

1. A method of setting a value r for a token rate parameter of an access regulator, wherein said access regulator controls admission of information from a communication device to a network and wherein said network comprises a node, the method comprising the steps of: receiving a signal at said access regulator from said node, said node comprising a buffer characterized by a buffer content level, said signal indicating the buffer content level of said node relative to a target level x, and setting said value r for said token rate parameter at a time (t+.DELTA.) based on a difference of a first value based on the rate r(t) at time t and a second value based on said signal indicating the relative buffer content, respectively, wherein said first value corresponds to the rate r at time t scaled by the complement of damping parameter .gamma. as scaled by time difference .DELTA. and said second value corresponds to said relative buffer content scaled by gain parameter a and the time difference .DELTA., wherein the damping and gain parameters .gamma. and a substantially satisfy the expression where .gamma. = .omega. tan .omega. and ~ is a propagation delay between said node and said access regulator.

2. In a network comprising a node and a set of K information sources, wherein the k th source in said set of K information sources is connected to a respective k th access regulator, wherein each access regulator is connected to said node, and wherein the k th access regulator is characterized by a rate, r k (t), at which information from the k th source is admitted to said network by said k th accessregulator, a method comprising the steps of:
receiving a signal at each access regulator from said node, said node comprising a buffer characterized by a buffer content level, said signal indicating said buffer content level relative to a target value; and setting the rate r k (t) of said k th access regulator at a time (t + .DELTA.) based on a difference of a first value based on the rate r k (t) at time t of said k th access regulator and a second value based on said signal indicating the relative buffer content, respectively, wherein said first value corresponds to the current rate r(t) scaled by the complement of damping parameter ~ as scaled by time difference .DELTA. and said second value corresponds to said relative buffer content scaled by a gain parameter a k and the time difference .DELTA., wherein the damping and gain parameters ~ and a k substantially satisfy the expression where ~ = .omega. tan .omega. and ~ is a propagation delay between said node and said access regulator.

3. The method of claim 1 wherein the damping and gain parameters ~ and a, and the target level ~ further satisfy the expression where

4. The method of claim 1 wherein values of the damping and gain parameters and a, and the target level ~ are such that is a value between a committed transmission rate and a substantially greatest realizable transmission rate.

5. The method of claim 1 wherein the damping and gain parameters ~ and a k, and the target level x further satisfy the expression where

6. The method of claim 2 wherein the damping and gain parameters ~ and a k, and the target level ~ are such that is a value between a sum of committed transmission rates and substantially greatest realizable transmission rates for said access regulator through said node.

7. The method of claim 2 wherein the rate r k (t) for the k th access regulator is proportional to the gain ratio