KR20010070121A - Method for fair flow control in packet-switched networks - Google Patents

Abstract

PURPOSE: A method is provided to be capable of securing minimum cell loss and maximum link utilization degree regardless of reciprocation time delay of an ABR(Available Bit Rate) loop. CONSTITUTION: A packet switched network includes a plurality of interconnected nodes(4,6,8,10), and each node connects a plurality of sources for transferring and receiving data and a data queue for storing data transferred from the sources. The data queue is related to current queue length and target value. Each node estimates the number of local bottleneck virtual lines using a minimum transfer rate to be secured and a data transfer rate from a corresponding source. The data transfer rate is allotted to each source based on a differential value of the current queue length, a difference between the current queue length and target value, and the estimated number of local bottleneck virtual lines, and the data transfer rate is allotted through a feedback signal transferred to each source.

Description

Method for fair flow control in packet switched network {METHOD FOR FAIR FLOW CONTROL IN PACKET-SWITCHED NETWORKS}

The present invention relates to a packet switched network, and more particularly to a method for fair flow control.

The packet switched network includes an asynchronous transfer mode (hereinafter referred to as an "ATM") network and the Internet, and fair flow control is a very important factor in transmitting information on a network. In particular, fair flow control in ATM networks is associated with Available Bit Rate ("ABR") services.

ATM typically achieves network use for information delivery through the following four service schemes. That is, the CBR (Constant Bit Rate) method that enables a fixed transmission bandwidth service, VBR (Variable Bit Rate) method that enables a variable bandwidth service while guaranteeing a constant cell loss rate, UBR (Unspecified Bit that does not determine the transmission bandwidth) Rate) method and ABR method. The ABR method changes the transmission bandwidth that a source can send according to network conditions. That is, the source can send data according to the available transmission bandwidth of the network.

ABR services were introduced in ATM networks to support data applications that could not be efficiently supported by bandwidth guarantee services such as VBR. Examples of thesis or data on this are as follows. S. Sathaye, "ATM Forum Management Specification, Version 4.0", Feb. 1996, and F. Bonomi and KW Fendick, "The Rate-Based Flow Control Framework for the Available Bit Rate ATM Service", IEEE Network 9 (2) (1995). p25-39, and R. Jain, "Congestion Control and Traffic Management in ATM Networks: Recent Advances and A Survey", Computer Networks and ISDN Systems 28 (13), (1996) p1723-1738.

Most data applications are very irregular and have no way of predicting future data traffic demands, while they have well-defined cell loss requirements and can have tolerances for time changes and unpredictable cell delays. More importantly, because of this feature, most data applications can change the data rate based on network load conditions. Accordingly, the concept of elastic traffic service has been introduced that adjusts the input data rate (or transmission flow) to the available bandwidth in the network. A representative example is ABR service in ATM network.

In principle, the ABR service does not require complex traffic characterization and call admission control on each of the sources and switches. By this simplicity, it is expected that implementation and development of ABR service will be much easier than bandwidth guarantee service (CBR or VBR service). However, the design and implementation of ABR-enabled switches has been determined to be far more difficult than originally anticipated. The biggest challenge is to design a scalable, stable and fair ABR flow control algorithm. In particular, the design of explicit rate (hereinafter referred to as "ER") allocation algorithm in asynchronous and distributed network environments.

The ATM Forum adopted a closed-loop rate-based approach for flow control of ABR services. Rate-based flow control, as implied by its name, uses information fed back from the network to control the rate at which each source can transmit to the network. The feedback information is delivered by a specific control cell called a resource management (hereinafter referred to as "RM") cell. There are three mechanisms for recording network congestion on the RM cell: explicit forward congestion indication (EFCI), relative rate (RR) marking, and explicit rate (ER) marking. Must be implemented on the switch.

On the other hand, the long and varied round-trip delay included in the flow control closed loop and the bottleneck located differently between the ABR VC (Virtual Circuit) make the design of high performance ER allocation algorithm difficult. ABR queues in a network are difficult to stabilize when the transmission rates of ABR sources are determined according to network state information at different points in time. In particular, using only the binary feedback mechanism (EFCI marking or RR marking, or EFCI marking and RR marking), ABR queues have persistent oscillation at steady state, and their amplitude is multiplied by the product of round trip time delay and available bandwidth. It will increase in proportion. A detailed description thereof is disclosed in the following example articles. E. Hernandez-Valencia et al., "Rate Control Algorithms for the ATM ABR Service", European Transactions on Telecommunications 8 (1) (1997) p7-20, and F. Bonnomi, D. Mitra and JB Serry, "Adaptive Algorithms for Feedback-Based Flow Control in High-Speed, Wide-Area ATM Networks, IEEE J. Select.Areas on Communications 13 (7) (1995) p1267-1283, and KK Ratmarkrishnan and Jain, "A Binary Feedback Sheme for Congestion Avoidance in Computer Networks with a Connectionless Network Layer, Proc. ACM SIGCOMM'88 (1988) 303-313.

This continuous vibration of the ABR queue generates periodic buffer offer and underflow, increasing cell loss potential and low link availability. The main premise of ER marking, ie the introduction of ER flow control, is to suppress the continuous vibrations generated by the binary feedback mechanism by ensuring asymptotic stability of the ABR queue. However, it is still difficult to design a simple form of the ER allocation algorithm that gradually stabilizes the ABR queue. This problem results in a feedback control problem with time delay from a mathematical point of view.

In their initial paper, Benmohanmed and Meerkov formulated the rate-based flow control problem as a discrete-time feedback control problem, and proposed an ER assignment algorithm that ensures asymptotic stability and enables arbitrary control of closed-loop performance. Benmohanmed and Meerkov's papers for which the ER algorithm is proposed are described in L. Benmohamed and S. M. Meerkov, "Feedback Control of Congestion in Packet Switching Networks: The Case of Single Congested Node", IEEE / ACM Trans. on Networking 1 (6) (1993) p693-708, and L. Benmohamed and SM Meerkov, "Feedback Control of Congestion in Packet Switching Networks: The Case of Multiple Congested Nodes", International Journal of Communication Systems, 10 (5) ( 1997) p227-246.

The ER allocation algorithm proposed by Benmohanmed and Meerkov is shown in Equation 1 below.

Where r [k] is the ER calculated by the switch at discrete time k, q [k] is the ABR queue length at discrete time k, and q _T is the desired queue length. And Is the controller gain, Is the maximum round trip time delay of ABR VC and I is any constant greater than "0". Despite the theoretical basis described in the above paper, the above algorithm is very complex in implementation and has a limited practical use. A. Kolarov and G. Ramamurthy, "A Control Theoretic Approach to the Design of close Loop Rate Based Flow Control for High Speed ATM Networks, Proc. IEEE INFOCOM'97 1 (1997), p293-301, describes this difficulty. As can be seen from Equation 1, the ER allocation algorithm is To time (j = 0 to It is required to maintain the value of the ER term of C) and requires a large number of floating point multiplications for every discrete time slot.

Whereas S. Chong, "Second-Order Rate-Based Flow Control with Dynamic Queue Threshold for High-Speed Wide-Area ATM Networks", preprint 1997, and A. Elwalid, "Analysis of Adaptive Rate-Based Congestion Control for High" Speed Wide-Area Networks, Proc. IEEE ICC'95 (1995) P1948-1953 proposed another ER allocation algorithm based on control theory in a simpler form, as shown in Equation 2 below.

In the first paper, we proposed the necessary and sufficient conditions for the closed-loop system to asymptotically stabilize when the algorithm of Equation 2 is applied. It extends to the general case with round trip time delay.

On the other hand, S. Chong, R. Nagarajan and YT Wang, in the paper "Designing Stable ABR Flow Control with Rate Feedback and Open-Loop Control: First-Order Control Case", Performance Evaluation 34 (4) We proposed a much simpler ER allocation algorithm. Regarding the ER allocation algorithm of Equation 3, disclosed in US Patent No. 5,864,538 issued January 26, 1999 under the name of the invention of "First-order rate-based flow control with adaptive queue threshold for ATM networks." It is.

r (t) = [-K` (q (t)-q _T )] ⁺ , K> 0

Here, [x] ⁺ = max [x, 0], which means to select a larger value between "x" and "0".

In the first of two papers related to Equation 2, two different stabilization conditions are derived. One of them is sufficient condition for the general case with heterogeneous round trip time delay, and the other is necessary sufficient condition for special case with homogeneous round trip time delay.

Compared to Equation 1, a common drawback of the ER algorithms disclosed in Equations 2 and 3 is that if the controller gain and queue length thresholds are available for the ABR traffic, then the available bandwidth used by the remote bottleneck VCs. If not properly selected according to instantaneous recognition in fraction, the ABR queue length may undesirably converge to "0" because the link is not sufficiently available at the equilibrium point. By definition, remote bottleneck VCs on a link are VCs that cannot be divided equally on that link because their transmission rate is limited by their PCR (Peak Cell Rate) or because bottlenecks occur on other links in the path. Conversely, if you apply an algorithm like Equation 1, there is no such thing as the undesirable equilibrium point mentioned above.

Accordingly, an object of the present invention is to provide a method for guaranteeing maximum link utilization and minimum cell loss irrespective of a round trip time delay of an ABR loop.

Another object of the present invention is to provide a method of minimizing the requirement of ABR queue size by ensuring the asymptotic stability of the ABR queue.

It is still another object of the present invention to provide a method for ensuring fairness of bandwidth utilization among ABR users, in particular, maximum-minimum fairness, which is an ATM forum standard.

It is still another object of the present invention to provide a method capable of rapidly adapting to a change in communication network environment, such as a change in the number of ABR users and a change in the ABR bandwidth.

It is still another object of the present invention to provide a method for performing almost all functions proposed in the ATM forum traffic management standard including functions such as EFCI marking, RR marking, and ER marking.

It is yet another object of the present invention to provide a method of having an asymptotic stabilizing operating point to achieve high availability, low cell loss, and MAX-MIN fair rate allocation.

It is still another object of the present invention to provide a method for improving responsiveness and control transient performance for multiple time scales, i.e., changing cell level rates of VBR and ABR VC and cell load arrival and departure of VBR and ABR VC. To provide.

It is yet another object of the present invention to provide a method for minimizing the number of operations required to compute an algorithm and reducing the complexity of the implementation and scaling to eliminate VC operations, including VC queuing, VC calculations, and VC table accesses. To provide.

In accordance with the above object, the present invention includes a plurality of nodes connected to the packet switching network, each node is connected to a plurality of sources for transmitting and receiving data, and a data queue for storing the data transmitted from the source And wherein the data queue is related to a current queue length and a target value, the method for fair flow control in the packet-switched network, wherein each node has a data rate in the corresponding source and a minimum data rate to be guaranteed. Estimating the number of local bottleneck virtual circuits using the method; and based on the difference between the current queue length and the target value, the derivative value of the current queue length, and the estimated number of local bottleneck virtual circuit lines, respectively. Assigns a data rate to the data source, wherein the data rate is transmitted to each of the sources. It characterized by constituted by any process that is allocated through a feedback signal.

1 is a network diagram for explaining the ABR service.

2 is a view for explaining each field included in an RM cell.

3 is a diagram showing a configuration having input and output port cards connected to each input and output port of the switch fabric for the ABR service.

4 is a detailed block diagram of the input / output port card of FIG. 3;

5 is a block diagram of an ABR service engine according to an embodiment of the present invention.

6 shows a network model with nodes of interest.

7 is a view showing a stable region in terms of U and V according to an embodiment of the present invention.

8 shows the relationship between the functions U, V and asymptotic damping rate α.

9 illustrates queue length thresholds for RR marking and ER marking.

10 is a peer to peer model configuration diagram.

11A-11D are experimental results in a peer-to-peer model configuration with only ER marking and no VBR background traffic.

FIG. 11A is a source transfer rate a _i (t) of VC with a 150 Mbps PCR, FIG. 11B is a source transfer rate a _i (t) of a VC with a PCR of 25 Mbps, FIG. 11C is a queue length at switch SW1, and FIG. Estimated number of local bottlenecks VC on switch SW1 to be.

12A to 12D are experimental results in a peer-to-peer model configuration with ER marking and VBR background traffic. FIG. 12A is a source transmission rate a _i (t) of VC with 150 Mbps of PCR, and FIG. 12B is 25 Mbps with PCR. Source transfer rate a _i (t) of VC, FIG. 12C is the queue length at switch SW1, and FIG. 12D is an estimate of the number of local bottlenecks VC at switch SW1. to be.

13 is a parking lot model configuration diagram.

14A to 14F are experimental results in a parking lot model configuration with only ER marking and no VBR background traffic. FIG. 14A is a source transfer rate a _i (t) of VC with 150 Mbps of PCR, and FIG. 14B is 25 Mbps with PCR. Source transfer rate a _i (t) of VC, FIG. 14C is the queue length at switch SW3, FIG. 14D is the queue length at switch SW4, and FIG. 14E is an estimate of the number of local bottlenecks VC at switch SW3. 14F is an estimate of the number of local bottlenecks VC in switch SW4. to be.

15A to 15F are experimental results of a parking lot model configuration with ER marking and VBR background traffic. FIG. 15A is a source transmission rate a _i (t) of VC having a 150 Mbps PCR, and FIG. 15B is a VC having a 25 Mbps PCR. Source transfer rate a _i (t), FIG. 15C is the queue length at switch SW3, FIG. 15D is the queue length at switch SW4, and FIG. 15E is an estimate of the number of local bottlenecks VC at switch SW3. 15F is an estimate of the number of local bottlenecks VC in switch SW4. to be.

16 is a control flowchart showing the ER allocation control operation performed by the ABR service engine according to an embodiment of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the same elements in the figures are represented by the same numerals wherever possible. In addition, detailed descriptions of well-known functions and configurations that may unnecessarily obscure the subject matter of the present invention will be omitted.

1 is a network diagram illustrating an ABR service. Referring to FIG. 1, a network 2 for an ABR service includes a plurality of nodes 4, 6, 8, and 10 connected to each other. The plurality of nodes 4, 6, 8, 10 are conceptual representations of the switches (hereinafter, nodes are also referred to as "switches"). Each of the plurality of switches 4, 6, 8, and 10 may be connected to a plurality of sources. In Fig. 1, sources A and C 12 and 14 are connected to the switch 4, and sources B and D 16 and 18 are connected to the switch 8, respectively. Each source 12, 14, 16, 18 transmits and receives data to the correspondingly connected switch 4,8. The data sent by the source is delivered to the destination via a path called a VC path containing multiple nodes. For example, data sent from source A 12 is delivered to source B 16 as the destination via the VC path where switches 4, 6 and 8 are present.

In the ABR service, a special cell for a control called an RM cell (denoted as RMc in FIG. 1) is provided to inform the source of available transmission bandwidth to the network. The generation of RM cells in the ABR service occurs at the sources 12, 14, 16 and 18 or the switches 4, 6, 8 and 10. In the present invention, only the RM cells generated by the source will be described.

The RM cell generated from the source passes through the switches 4, 6, and 8 via the VC path to the destination, and this cell traveling direction is called forward. At the destination, the received RM cell is sent back to the source. The forward direction of the RM cell returned to the source is called backward. In Fig. 1, the source is assumed to be 12,14 and the destination is assumed to be 16,18 to indicate the forward and reverse directions. At this time, each switch (8, 6, 4) stores the information of the allowable bandwidth in the RM cell and the stored information is transmitted to the source (12,14) so that the source (12,14) referring to this information Determines by appropriately adjusting the amount of information sent to destinations 16 and 18.

The RM cell contains a lot of information, among which the most important are the CCR (Current Cell Rate), MCR (Minimum Cell Rate), ER (Explicit Rate), NI (No Increase), and CI (Congestion Indication) fields. 2 illustrates fields included in the RM cell. Referring to FIG. 2, a source-receive (S-R) field is a field for storing source and destination related information, and a CCR field is a field for storing a transmission bandwidth value of a source when the source first generates an RM cell. The MCR field is a value stored when a source generates an RM cell, and indicates a minimum transmission rate of each VC. The ER field is a field used in the reverse direction in which the RM cell generated by the source passes through the forward path and then returns to the source again. Each switch in the reverse direction writes its available bandwidth information to the RM cell, where the information is stored in the ER field in the RM cell. When the RM cell arrives at the switch, the stored method stores the smaller value among the available bandwidth already stored in the ER field and the available bandwidth calculated by the switch as the new value of the ER field. That is, the source receives the ER value calculated at the switch having the smallest available bandwidth among the paths of the VC. To do this, the source that first generates the RM cell will use the peak bandwidth (PCR), the maximum bandwidth it can send to the ER field. In addition, NI and CI are fields used for relative rate control. The NI field is a field for informing the source that switches on the path through the VC do not increase the transmission bandwidth of the source according to the traffic congestion level. This field is used to inform the source that, in the worst case, the source should reduce its transmission bandwidth.

For ABR service, each switch needs to write the amount of transmission bandwidth allowable in the ER field in the RM cell when the reverse RM cell passes. To do this, each switch must calculate the available bandwidth and a switch algorithm is required to calculate it. In the end, what we want to get through the switch algorithm is the value of the transmission bandwidth that the switch can allow for each VC. This value of the transmission bandwidth is known to the source, and the source adjusts the transmission rate with reference to this value, thereby providing a normal ABR service.

For the ABR service, each switch 4, 6, 8, and 10 of FIG. 1 is provided with an input / output port card 22 in each of the input / output ports of the switch fabric 20 as shown in FIG. As illustrated in FIG. 4, the input / output port card 22 includes an input / output buffer manager 30, an ABR service engine 32, and an output interface 34. The input / output buffer management unit 30 is connected to the switch fabric 20 and manages the input / output queuing. The input / output buffer management unit 30 includes an ABR queue 36. The output interface 34 performs the user network interface function of the ATM layer, and the ABR service engine 32 processes the ABR algorithm and related functions according to an embodiment of the present invention while the RM cell passes through it.

An example of a concrete block configuration of the ABR service engine 32 of FIG. 4 is shown in FIG. 5. Referring to FIG. 5, the ABR service engine 32 includes an EFCI marking unit 40, an Q estimator 42, an ER calculator 44, a queue average calculator 46, and a reverse RM cell recorder 48. ) Is included.

In FIG. 5, the explicit forward congestion indication (EFCI) marking unit 40 marks an EFCI bit for notifying the EFCI congestion to the forward data cell applied when the EFCI congestion occurs. A cue read signal and a cue write signal are input to the queue average value calculator 46 of FIG. The cue read signal is a signal applied from the outside when the cell is removed from the queue, and the cue write signal is a signal applied from the outside when a new cell is input to the queue. The queue average value calculator 46 increases the value of the instantaneous variable of the queue when the cue write signal is applied, and decreases the value of the instantaneous variable of the queue when the cue read signal is applied, thereby being used in calculating the ER value. Calculate the average of the cue values during the calculation cycle. The queue average calculator 46 calculates an average of the calculated queue values. Is output to the ER calculator 44. In addition, the queue average calculator 46 determines the EFCI congestion by using the cue read signal and the cue write signal, and outputs a signal EFCI_CG informing the EFCI congestion to the EFCI marking unit 40 where the EFCI congestion has occurred. In addition, according to the embodiment of the present invention, the congestion based on the low threshold q _LT and the high threshold q _HT preset in the ABR queue and the signal CG (congested) and VCG (very_congested) according to the degree of congestion are reverse RM cell recording unit 48. ) The congestion and very congestion signals CG and VCG will be used for relative rate control according to an embodiment of the present invention.

In Fig. 5, the | Q | estimator 42 is a block for estimating the number of local bottleneck VCs. The Q estimator 42 reads the CCR and MCR values in the RM cell, compares the periodically calculated ER value r (t), and estimates the number of local bottleneck VCs. Calculate Estimate The interval W for calculating s is described in detail in the following section ((5) Discrete-Time ER Algorithm and Q Estimation). The estimate Is provided to the ER calculator 44.

In FIG. 5, the ER calculator 44 periodically calculates and updates the ER value and outputs the most recently updated ER value r (t) to the reverse RM cell recorder 48 when the reverse RM cell arrives.

In FIG. 5, the reverse RM cell recording unit 48 records the ER value r (t) periodically calculated by the ER calculator 44 in the ER field in the RM cell. More specifically, the reverse RM cell recording unit 48 adds the ER value in the current RM cell and the ER value r (t) provided by the ER calculator 44 plus the MCR value in the RM cell (the specification of the present invention). R (t) + mi), and writes r (t) + mi in the ER field in the RM cell only when r (t) + mi is smaller. This is to perform explicit transmission rate control. In addition, the reverse RM cell recording unit 48 records the binary logic bits in the NI field and the CI field in the RM cell according to the congestion and very congestion signals CG and VCG related to the relative rate control provided by the queue average value calculator 46. do. This is to perform relative rate control.

The ER allocation control operation performed by the ABR service engine 32 of FIG. 4 will be briefly described with reference to the control flowchart of FIG. 13 and the concrete block configuration of FIG. 5.

The ABR service engine 32 estimates the Q value by the Q estimator 42 in step 100 and proceeds to step 102. In step 102, the ER calculator 44 updates the periodic ER calculation and the calculated ER value r (t). After that, when the RM cell arrives at the reverse RM cell recorder 46 (step 104 in FIG. 7), the ABR service engine 32 proceeds to step 106 in FIG. 7. In step 106, the most recently updated ER value r (t) is read by the reverse RM cell recorder 46, and the MCR value mi is read from the received RM cell. Thereafter, in step 108, the ER allocation value ri (t) = r (t) + mi of the VCi is calculated by the reverse RM cell recording unit 46. Thereafter, in step 110, the calculated ri (t) is recorded in the ER region of the reverse RM cell.

The proposed switch algorithm according to an embodiment of the present invention aims to achieve a convergence value of the queue state of the ABR queue 36 and to fix the queue length value in a steady state to a target value q _T. Keeping the length of the queue constant includes the fact that the amount of input and output traffic entering the ABR queue 36 is equal. If up to a queue length value q _T to reach the target there is a transient state so that the lift amount of the example input traffic standing compared to output the traffic volume may momentarily considerably larger at a time can be greatly requires ABR queue 36. This leads to unexpected cell loss. Therefore, the embodiment of the present invention implements a switch algorithm considering the change in the buffer amount of the ABR queue 36 and the queue length state (buffer state).

The embodiment of the present invention proposes a continuous time ER allocation algorithm as shown in Equation 4 to have a form of a controller capable of removing the undesired equilibrium point in the ABR queue 36 as simply and simultaneously as possible.

Where r (t) is the calculated ER value, Is the derivative of r (t). Q is a set of locally bottle-necked VCs in the link, and Q is the cardinality of Q. q (t) is the length of the ABR queue 36 at time t, and q_T is the length (ie, the target value) of the target ABR queue 36. A and B are the gains of the controller. In the embodiment of the present invention, the values A and B are set differently according to the length of the ABR queue 36. This setting causes asymptotic stability.

Comparing the ER allocation algorithm according to the embodiment of the present invention represented by Equation 4 with the algorithm shown in Equation 2, the algorithm represented by Equation 4 according to the embodiment of the present invention is applied to the damping term (right side claim 1). instead of r (t) Note that we use. Is a derivative value of q (t), which will be described in detail with reference to Equations 21 and 23, which will be described later.

This change in an embodiment of the present invention eliminates undesirable equilibrium in closed loop systems, resulting in bandwidth and remote bottle-necked VCs available to the ABR queue 36. regardless of the portion of available bandwidth is used by the always converge to a target value _T q. This means that full use of the available bandwidth at steady state is always guaranteed.

Another notable feature of an embodiment of the present invention is that the controller gains A and B are normalized by the number of local bottleneck VCs. This normalization allows the asymptotic damping rate of the closed loop system to be independent of the number of local bottleneck VCs. Estimation of the number of local bottleneck VCs | Q│ has been an important topic in the rate-based ABR flow control study. These papers include M. K. Wong and F. Bonomi, "A Novel Explicit Rate Congestion Control Algorithm", it Proc. IEEE GLOBECOM'98 4 (1998) p 2432-2439 and L. Kalampoukas, A. Varma and KK Ramakrishnan, "An Efficient Rate Allocation Algorithm for ATM Networks Providing Max-Min Fairness", Technical Report UCSC-CRL-95-29 , Computer Engineering Dept., University of California, Santa Cruz, June 1995, and A. Charny, KK Ramakrishnam and A. Lauck, "Time Scale Analysis and Scalability Issue for Explicit Rate Allocation in ATM Networks", IEEE / ACM Trans. on Networking 4 (1996) p569-581 and R. Jain et al., "ERICA Switch Algorithm: A Complete Description", ATM Forum / 96-1172 (1996).

The difficulty that lies in the dynamic nature of the estimation process is the dynamic nature of the ER-allocated processor, that is, the ER update affects the QQ update or vice versa until the closed-loop system reaches steady state. As a result, the inadequate design of the | Q | estimation algorithm can result in unstable closed-loop systems. A robust, stable and measurable estimation algorithm is provided in relation to the ER algorithm according to an embodiment of the present invention.

According to an embodiment of the present invention, the control theory ER algorithm represented by Equation 4 ensures the so-called MAX-MIN fair rate allocation, for the following reason. All VCs based on the calculation of Equation 4 at the switch always assign the same ER (hereinafter referred to as "common ER") to the shared link. Therefore, the RM cell transmits the minimum allocated ER through the path to the source, and the source transmits data to the minimum ER. The ABR queue 36 at the switch will have a delay and attenuate if the VC is bottlenecked elsewhere and the source transmits data at a lower rate than the common ER allocated by the switch. If the ABR queue 36 falls below the target queue length q _T , the switch increases the common ER to be the target queue length q _T according to the algorithm shown in equation (4). So the local bottleneck VCs will eventually use the same unused bandwidth by the remote bottleneck VCs.

Indeed, algorithms such as Equation 4 should be replaced by discrete time concepts if applied to a network operating situation. The algorithm of Equation 4 presented in accordance with an embodiment of the present invention is represented by Equation 5 below.

Where T is the duration of the discrete time slot. The algorithm of equation (5) is a special case of the algorithm of equation (1). That is, I = 1, , And = 0, Substituting and converting the result becomes an algorithm of Equation 5. In contrast to the algorithm of Equation 1, the converted algorithm no longer allows any control of the closed loop dynamics. However, the inventors of the present invention argue that having arbitrary control capability is, as discussed, very expensive in terms of implementation cost and not required for ABR flow control design. To support this assertion, embodiments of the present invention will demonstrate that the converted algorithm can accept control over closed loop performance.

In the control theory ER algorithm (Equation 4 and Equation 5) according to an embodiment of the present invention, queue length control is of primary interest and fair rate allocation is a secondary result of queue length control. In other forms of ER algorithms (14) to (18) described below, maximum-minimum fairness is of primary concern and queue length control is secondary. The other type of ER algorithm tracks the available bandwidth for ABR traffic and requires each switch to share a fair share of remote bottleneck VCs in the number of local bottleneck VCs and its own bottleneck link of VCs. Based on this information, the switch updates per-VC ER allocation in a way that achieves MAX-MIN fair rate allocation and target link utilization incrementally.

(1) ER rate flow control

In the following, the ER allocation algorithm of the present invention is mathematically modeled, and through the modeling, it will be explained that the ER allocation algorithm of the embodiment of the present invention satisfies the maximum-minimum fairness while ensuring the MCR required for each source.

6 shows a network model with nodes of interest. In FIG. 5, each of the plurality of VCs includes a plurality of corresponding sources. That is, the VCi 54 corresponds to the source i 50 and the VCj 56 corresponds to the source j 52. 36 is an ABR queue, q _T indicated on the ABR queue 36 is a target value of the ABR queue 36, and is connected to the ABR queue 36. Is the available bandwidth of the shared link of the plurality of VCs. Also And Each means a forward path delay and a reverse path delay in VCi 54, Wow The sum of the delays is the round trip time delay in VCi 54. Becomes And Each means a forward path delay and a reverse path delay in VCj 56, And Sum of round trip time delay in VCj 56 Becomes

First, assume the following for the analysis of the network model.

Assumption 1) Traffic is considered a determined flow flow, and network queuing processors and flow control mechanisms are considered to be continuous in time. This assumption makes the model a closed loop system by differential equations.

Assumption 2) Round-trip delay of VCi Is the forward path delay, including propagation, queuing, transmission, and processing time And reverse path delay Sum of Here, it is assumed that the round trip time delay is constant.

Assumption 3) Sources have persistent characteristics. Here, continuous means that the source can send enough data at the rate assigned to the network.

Assumption 4) There is no arrival and departure of VCs until the system reaches steady state.

Assumption 5) Available Bandwidth at the Link Is constant until the system reaches steady state. Also assume that the buffer size of the ABR queue 36 at the link is infinite.

The transfer rate when the source i 50 transmits data at source time t is denoted by a _i (t), and the ER value allocated to VCi 54 as the node of interest calculates the ER at node time t. We write r _i (t). In addition, the most recent minimum value of the ER value assigned to the VCi 54 by the node is denoted by b _i (t), and the peak rate constrained by the VCi 54 is p _i (t). Mark it. P _i (t) is a Peak Cell Rate (PCR) in the VCi 54 as a standard term in ATM ABR.

Ignoring the linear increase and exponential decrease based on the binary feedback, the operating characteristics of the source are modeled as in Equation 6 below.

Where N is the set of all VCs including the node of interest to the root. In the model shown in Equation 6, source i 54 is a reverse path delay. The ER value previously assigned by the node of interest This means that data is transmitted at the lowest transmission rate among the minimum ER value, i.e., bi (t) and VC PCR, i.

On the other hand, the operation characteristics of the class-specific ABR queue 36 of interest in the node are given by Equation 7 below.

An ER allocation algorithm according to an embodiment of the present invention has a queue length q (t) and a derivative of the queue length. , Estimate of the number of local bottleneck VCs Algorithm to perform in each switch based on the current network state, including. This switch algorithm is given by the equations (8) and (9).

In Equation 9, the controller gains A and B are A, B > 0. M _i in Equation 8 is a minimum data rate at a node for which source i 50 guarantees during the life of VCi, and MCR (Minimum Cell Rate) is a standard term in ATM ABR. here Assume that, as expressed in Equation 10, call admission control is guaranteed.

r (t) is the ER allocation value per VC It should be noted that the common part of, i.e., common ER value, is the most needed for algorithm calculation. In the calculation for each VC required according to an embodiment of the present invention, only the common ER value r (t) is added to m _i, which is an MCR value guaranteed by the source i 50. This simple operation allows the embodiment of the present invention to reduce the complexity of the calculation with an increase in the number of VCs and to have scalability. At present, each RM cell of the ABR service supporting VCi 54 carries an MCR value m _i in a round trip.

According to an embodiment of the present invention, the common ER value r (t) is updated by "background calculation". "Background calculation" means that the node calculates the common ER value periodically regardless of the RM cell reception. An advantage of performing the background calculation as described above is that when the RM cell arrives at the corresponding node, it is possible to immediately provide the most recently updated common ER value r (t).

Node sewn to update a common ER value r (t) obtained in accordance with Equation (9) through the background calculated in MCR field of the RM cell in VCi (54) through reading the m _i, updates the read m _i and the last By substituting the obtained common ER value r (t) into Equation 8, the ER value r _i (t) to be assigned to the VCi 54 is obtained, and the value r _i (t) is recorded in the ER field of the RM cell.

Another notable feature of the algorithm according to the embodiment of the present invention is an estimate of the number of local VCs. This is in the normalization of the controller gains A and B. This normalization is not absolutely necessary and is recommended. The normalization allows the asymptotic decay rate of the Peruvian system to be independent of the number of local VCs. The normalization of the controller gains A and B will be described in detail in the following "(4) Main Root and Asymptotic Attenuation Rate".

The remote bottleneck VC and the local bottleneck VC are defined at steady state for a preset network load. Remote bottleneck VCs on a link are defined as VCs that cannot be shared equally on that link either due to limited throughput by PCR or because of bottlenecks on other links in the path. A local bottleneck VC on a link is defined as VCs that can be shared fairly on that link.

To mathematically define local bottleneck VCs and remote bottleneck VCs, , , In other words, the set Q of all local VCs is given by Equation 11 below. Here, a _is, r _{is, and} b _is means a _i (t) _, r _i (t), and b _i (t) at steady state.

The set N-Q of all remote VCs is given by Equation 12 below.

| Q | in the "(5) Discrete-Time ER Algorithms and | Q | Estimate" chapter below In the estimation algorithm, the estimate Fast convergence to | Q | is not efficient, but It will be described in detail that R has a greater robustness than | Q |. The stiffness characteristic is necessary for the stability of the ABR flow control.

(2) steady state and impartiality

In the following, steady state characteristics of the ABR closed loop dynamic characteristics will be described when the ER algorithm according to the embodiment of the present invention is applied. That is, the analysis result according to the embodiment of the present invention will be described. Closed-loop dynamic characteristics indicate that all derivatives of system variables are "0" , , Assume that we have an equilibrium point of By these assumptions, the following equations (13), (14) and (15) are obtained from equations (6), (8) and (9).

,

r _is = r _s + m _i , ∀i ∈ N,

q _s = q _T

here, to be. Other notations have been defined in the preceding paragraph. Since q _s = q _T > 0, the buffer equation of Equation 7 has the same meaning as in Equation 16 below.

By combining the equations of equations (13), (14), and (16) with the definitions of equations (11) and (12), the following equation (17) can be obtained.

Equation 17 means the following Equation 18.

As described above, in the exemplary embodiment of the present invention, the steady state has the following characteristics, and the result as shown in Equation 19 is obtained.

For (i) the queue length is equal to the target queue length (q _s = qT), and (ii) the available bandwidth on the link is used as much as possible ( ), (iii) each MCR is guaranteed on each link, and the bandwidth deducted by the sum of the MCRs There is a steady state uniqueness (equilibrium point), which is shared equally in terms of MAX-MIN fairness. In other words,

As described above, when the ER algorithm according to the embodiment of the present invention is applied, the ABR closed-loop system has a unique and vibration-free operating point, and regardless of the network load at the operating point without vibration. In addition to MCR guarantees, maximum minimum (MAX-MIN) fairness is achieved. The queue length of the ABR queue 36 is equal to the target value q_T. That is, the equilibrium point is completely independent of the bandwidth available for ABR traffic and some of the available bandwidth used by remote bottleneck VCs, including PCR-bound VCs.

It is surprising that the ER algorithm of the embodiment of the present invention is simple while creating such a good equilibrium point. The ER algorithm according to an embodiment of the present invention is q (t) and Only information is required, no other measurement or monitoring results are required. Estimate Is used to normalize the controller gains A and B, but is introduced only to promote the convergence rate and has nothing to do with the equilibrium point. The Benmohamed's algorithm disclosed in Equation 1 also yields a unique system equilibrium point having the same characteristics as the algorithm of the embodiment of the present invention, but with a high implementation complexity. In contrast, if the algorithms disclosed in equations (2) and (3) are applied, the closed-loop system has an equilibrium point that varies based on the available bandwidth and the available bandwidth used by the remote bottleneck VCs. The queue length may converge to "0". This is described by S. Chong, R. Nagarijan and YT Wang, "Designing Stable ABR Flow Control with Rate Feedback and Open-Loop Control: First-Order Control Case", Performance Evaluation 34 (4) (1998) p189-206. , S. Chong, "Second-Order Rate-Based Flow Control with Dynamic Queue Threshold for High-Speed Wide-Area ATM Networks", preprint 1997, and A. Elwalid, "Analysis of Adaptive Rate-Based Congestion Control for High-Speed Wide-Area Networks, Proc. IEEE ICC'95 (1995) P1948-1953.

(3) asymptotic stability

Hereinafter, the stability of the ER algorithm according to an embodiment of the present invention will be described. In general, the stability characteristics of the equilibrium point given in Equation 19 above can be investigated in the case of multiple nodes. However, due to the complexity of the coupling dynamic characteristics between the nodes, the analysis may be too complex, so embodiments of the present invention do not attempt to solve the overall stability problem in the combined multi-node configuration. In the embodiment of the present invention, this is investigated only in the simulation in the following "(6) simulation results" chapter. On the other hand, in the paper L. Benmohamed and SM Meerkov, "Feedback Control of Congestion in Packet Switching Networks: The Case of Multiple Congested Nodes", international Journal of Communication Systems, 10 (5) (1997) p227-246, Node-to-node dynamics are separated, and local stability conditions derived from adjacent points of equilibrium are also available for FCFS service schemes. Depending on these results, it is assumed that adjacent points having the property R also exist in the case of the present invention. Consider a subset of R including equilibrium points. Here, the followings are satisfied.

a) That is, the dynamic characteristics of the remaining nodes are in a steady state.

b) And That is, the local bottleneck VCs To send data to the remote bottleneck VCs, Send data to

c) Saturation nonlinear values in Equations 7 and 8 are not activated. That is, q (t) and r (t) are positive values.

d) The estimation process is in a steady state. In other words, Is a constant.

In the vicinity of this equilibrium point, the dynamic characteristic equations (6), (7) and (9) can be simplified to the following equations (20), (21) and (22).

By combining Equations 20 and 21, the following Equation 23 can be obtained.

If the error function is defined as e (t) = q (t)-q _T , and the derivatives of Equations 22 and 23 and the derivatives of Equation 8 are combined, a closed loop equation can be obtained. have.

Equation 24 is the second order delay differential equation. The characteristic equation of the closed loop equation is given by Equation 25 below.

Equation 25 has an infinite number of roots. With respect to the asymptotic stability of the closed loop equation of (24), all roots of the characteristic equation of (25) should have negative real parts. This is described in R. Bellman and KL Cooke, "Differential-Difference Equations" (Academic Press, New York, 1963), and G. Stepan, "Retarded Dynamical Systems: Stability and Characteristic Functions" (Longman Scientific & Technical, 1989). Doing.

The equation of D (s) = 0 can rely on Fonttryagin's criterion, assuming discrete delays of rational ratios, to find the necessary full condition with stable roots. See R. Bellman and K. L. Cooke, "Differential-Difference Equations" (Academic Press, New York, 1963), and S. J. Bhatt and C.S. Hsu, Stability Criteria for Second-Order Dynamical Systems with Time Lag, Journal of Applied Mechanics "(1996) p113-118. In the general case involving continuous delays or discrete delays of flock ratios, stepan's criterion provides a way to derive the necessary and sufficient conditions, as described in G. Stepan, "Retarded Dynamical Systems: Stability and Characteristic Functions" (Longman Scientific & Technical, 1989). Configuring in a positive form is very complex in the case of multiple connections of various round trip delays.

Instead, embodiments of the present invention derive the necessary and sufficient conditions for asymptotic stability when all round trip time delays are the same. When τ_i = τ and ∀_i, the closed loop equation of Equation (24) is represented by the following Equation (26).

The equation of Equation 26 can be normalized so that the time delay τ is 1. If t = τξ, the equation of Equation 26 is represented by the new variable ξ as shown in Equation 27 below.

In the equation of Equation 27, , to be. The characteristic equation of Equation 27 is as follows.

The Fontrian criterion described above can be used to find the necessary sufficient condition that all roots of H (z) = 0 have a negative real part.

The above theory leads to the following conclusions. As mentioned above , Put it on,

The closed loop equation of Equation 26 has gradual stability only when the following inequality condition is satisfied.

In the two inequalities, ω_1 is the interval Is the only solution for U = ωsinω. In addition, the gains A and B of the controller can be determined by the above inequalities. 7 shows a stable region in terms of U and V according to an embodiment of the present invention.

(4) major root and asymptotic damping rates

In the following it will be described to determine the rate at which a stable closed loop system approaches a steady state according to an embodiment of the present invention. The solution of the normalized closed loop equation of Equation 27 can be represented by the same series as Equation 30 below.

Where p _n (ξ) is a suitable polynomial and z _n , ∀n are the roots of the corresponding characteristic equation of equation (28). As a root with the largest real part Consider the main root, denoted by. When released, the following equation (31) is derived from the equation (30).

Here, C is a constant depending on the initial condition of Equation 27, Is Indicates asymptotically approaching 1. In addition, the following equation (32) is derived from the equation (30).

From here, Represents the Euclidean norm, and c is a constant depending on the initial condition of equation (27). In equation 32, the circular variable t (= ) Can be expressed as in Equation 33 below.

here, It should be noted that the circular system is the asymptotic damping rate towards the equilibrium point. Inverse of the asymptotic attenuation rate Is the time constant of the circular closed loop system, Has a small perturbation around the equilibrium point It is the time taken to decrease by the factor of. Where α is the asymptotic damping rate, Is the time constant of the normalized system.

8 shows the relationship between the functions U and V and the asymptotic damping rate α. Referring to Figure 8, it is a concave function with a maximum of about 0.3 near point (U, V) = (0.6, 0.1). The contour line at α = 0 corresponds to the boundary of the stable region shown in FIG. 7. Once determined for a given (U, V) pair, β can easily be determined by substituting z = -α + jβ in the characteristic equation of Equation 28 and calculating the real and imaginary parts.

(5) a discrete-time ER algorithm and -calculation

So far, we have described the continuous-time model of the ABR flow control problem. However, since the feedback information is actually relayed through the RM cells, the sampled form is more useful than in continuous time. Moreover, since the RM cells themselves must compete to obtain link bandwidth along the round trip time path, the feedback information is not periodic and the time intervals of the RM cells are time-varying.

According to the exemplary embodiment of the present invention, the ER algorithms in Equations 8 and 9 are implemented in discrete time in the switch as shown in Equation 34 below. The common ER is periodically updated with a period of T by the following equation (34).

, A, B> 0

In Equation 48, Denotes the lowpass filtered queue length, that is, the queue average. In the embodiment of the present invention We use the period-average filter to find Becomes

It should be noted that the discrete time ER update equation of Equation 34 corresponds to Equation 9 when T → ∞. In contrast to the periodic calculation of the common ER values r [k], r [k + 1], ER value allocation per VC is performed aperiodically at the moment when the corresponding RM cells arrive in the forward or reverse direction. That is, when the RM cell of VC _i arrives at time t, the switch calculates the following equation (35).

r _i (t) = r (t) + m _i

The result according to Equation 35 is recorded in the ER field of the RM cell. In the equation of Equation 35, r (t) represents the latest value r [k + 1] of the common ER value r [k], and in the embodiment of the present invention, it is calculated periodically, that is, background Is updated. m _i is available from the arriving RM cell or from the MCR table per VC maintained in the switch and is updated at the start and arrival of the call. If the embodiment of the present invention decides to take the value of m _i from the arrived RM cell, the MCR table per VC and access to it are no longer needed. Then, only necessary operations per VC in the algorithm according to the embodiment of the present invention may be performed by addition as shown in Equation 35.

Meanwhile, many theoretical systems have been proposed for estimating or tracking local bottleneck VCs. An example of such a theoretical system is M. K. Wong and F. Bonomi, "A Novel Explicit Rate Congestion Control Algorithm", it Proc. IEEE GLOBELCOM'98 4 (1998) p2432-2439 and L. Kalampoukas, A. Varma and KK Ramakrishnan, "An Efficient Allocation Algorithm for ATM Networks Providing Max-Min Fairness", Technical Report UCSC-CRL-95-29, Computer Engineering Dept., University of California, Santa Cruz, June 1995, and A. Charny, KK Ramakrishnan and A. Lauck, "Time Scale Analysis and Scalbility Issue for Explicit Rate Allocation in ATM Networks, IEEE / ACM Trans. On Networking 4 ( 1996) p569-581, and R. Jain et al., “ERICA Switch Algorithm: A Complete Description”, ATM Forum / 96-1172 (1996), each of which differs in the degree of implementation complexity. C. F, Su, G. de Veciana and J. Warlrand, "ON" sources that share a link in an algorithm such as "Explicit Rate Flow Control for ABR Services in ATM Networks, preprint, 1997". The basic concept of estimating numbers is attractive because it does not require calculations per VC.

In an embodiment of the present invention, the algorithm is modified to estimate the number of local bottleneck VCs without calculation per VC.

Assume that the j ^th RM cell arrives at the switch at switch time t ^j . According to ABR rule [1], if j ^th RM cell is an RM cell of VC _i , the j ^th RM cell carries a _i (t ^j -τ _i ^f ) value of the CCR field and m_i value of the MCR field. . The switch monitors the arrival of the RM cell synchronously over fixed time intervals of W seconds. For the l ^th interval, the number of local bottlenecks VCs can be approximated by Equation 36 below.

In Equation 36, 1 {·} is an indicator function, CCR (t ^j ) means a value in the CCR field, and MCR (t ^j ) means a value in the MCR field. r (t ^j ) represents the latest value of the common ER value at time t ^j . In addition, NRM is a period in a cell unit that sends a forward RM cell and is negotiated at the time of establishing a VC connection. When the j ^ th RM cell arrives, if the current cell rate deducting the MCR is greater than or equal to the latest value of the common ER value at the switch, the VC to which the j ^th RM cell belongs is calculated as the local bottleneck VC. δ is in particular a limit value to prevent underestimation of the number of local bottleneck VCs near the steady state. When the system approaches steady state, the current cell rate of the local bottleneck VC converges to the sum of the MCR and common ER values, which would be incorrectly calculated as the remote bottleneck VC due to the small perturbation at the current cell rate without a limit δ. Can be. However, by having the above limit value δ, the above underestimation can be effectively avoided. As a simulation result in the embodiment of the present invention, δ = 0.9 is preferable. In addition, the expected number of RM cell arrivals in the VC within W seconds, i.e. Note that since the value of the indicator function is normalized by, the sum over these W second time intervals gives an accurate estimate of the number of local bottleneck VCs. An iterative estimate based on these estimates for each time interval is calculated at the following equations 37 and 38 at all time intervals.

In Equation 37, λ is an average factor, int [a] represents a minimum positive number of a or more, and a saturation function sat _a [b] is defined as in Equation 39 below.

Against all t It should be noted that The actual number of local bottleneck VCs may be "0" for any networking load, but in embodiments of the present invention intentionally estimated to avoid dividing by values close to "0". Leave the lower limit.

Now the question is how to choose the interval W and the mean factor λ. As the number of VCs sharing a link increases or the available bandwidth decreases for a given interval W and the arrival latency of the RM cells of the VC increases, the switch selects fewer RM cells of VC for the interval W. Is observed, and eventually | Q | The estimate of _l begins to change significantly. In order to solve this problem, it is also possible to adjust W according to the number of VCs sharing links and changes in available bandwidth. But this is not easy to implement. Instead, the periodic averaging operation in (37) effectively A large lambda can be chosen from a value close to 1, hoping to eliminate the variability of. In the embodiment of the present invention, through the simulation, a stable and effective with respect to the number and available bandwidth of the VCs sharing a wide range of one link The estimated value of was found to be λ = 0.98 regardless of the choice of the interval W.

The inventors of the present invention to increase the stability Improve the estimation algorithm further. First, assume that controller gains A and B are selected as in Equation 40 below.

And (U, V) pairs exist in the stable region shown in FIG. The estimate is accurate enough for all times The selection of the controller gains A, B is made to be true. However, in general Since the normalized closed loop equations, i.e., the actual (or effective) values U 'and V' of U and V that control Equation 27, are given by Equation 41 by substituting Equation 40 in Equation 26.

Consider the stable region shown in FIG. if If is less than 1, the points (U ', V') are on a straight line connecting the points (U, V) and the origin (0, 0), so they are in the stable region. Contrary, When this exceeds 1, the point (U ', V') moves up along a straight line containing the origin (0,0) and the point (U, V), and consequently leaves the stable region. The overestimation of can be allowed to change only the asymptotic damping rate without affecting the stability of the system, Underestimation of the system should be avoided because it makes the system unstable. The above is the reason why the limit value δ is necessary and introduced in Equation 36 above. However, δ does not solve all situations, especially in situations where the new ABR VC participates in a small initial cell rate (ICR (Initial Cell Rate) in ATM ABR terminology). If the new VC participates in a small initial cell rate, it is very likely to be counted as a remote bottleneck VC during the transient period. This is because, if the limit δ is not small enough, it is very likely that CCR (t ^j )-MCR (t ^j ) is smaller than δ · r (t ^j ) in (36). In this situation, if the new VC is actually a remote bottleneck VC this is not a miscalculation, but if the new VC is actually a local bottleneck VC The initial underestimation of can make the system diverge. In order to solve this problem, the embodiment of the present invention Improve the estimation algorithm as follows:

As soon as a new VC arrives at time t, it is intentionally updated as Equations 42 and 43 as if VC is a local bottleneck VC.

By doing as described above, Avoid underestimation of Is from (42) and (43). Converge on. This is a very desirable property for system stability as described above. In the chapter "(6) Results of Simulation", which will be described later, the simulations are proposed in relation to the ER algorithm according to an embodiment of the present invention. -Check the performance of the estimate.

Next, an embodiment of the present invention considers a case in which RR (relative rate) marking is applied in relation to ER marking. While ER marking plays an important role in achieving MAX-MIN fairness as well as MCR guarantees, RR marking plays a role in suppressing excessive overshoot of possible queue lengths (i.e., transient cells). Minimize losses).

9 shows a state where the queue length thresholds for these RR and ER markings are set. In FIG. 9, q _T is the desired queue length, q _LT is the low threshold of the ABR queue, and q _HT is the high threshold of the ABR queue 36. Wherein _T q is related to ER marking, the q _LT and _HT q is related to the RR marking. The threshold q _HT is a threshold for setting the CI bit in the CI field of FIG. 2, and q _LT is a threshold for setting the NI bit in the NI field of FIG. 2. If the queue length of the ABR queue 36 is greater than the threshold q _HT , the switch sets the CI bit of the CI (Congestion Indication) field in the reverse RM cell to "1" to indicate an excess of queue length. If the queue length of the ABR queue 36 is between the thresholds q _LT and q _HT , the NI bit of the No Increase (NI) field in the reverse RM cell is set to "1" to indicate that the source should no longer increase the transmission bandwidth. Set. In the embodiment of the present invention, by setting the target queue length q _T much lower than the low threshold q _LT , the queue length may remain near q _T under the control of the ER marking unless an unexpected change in network load occurs. RR marking is rarely active. If the cue length exceeds q _{HT due} to a sudden load change, in an embodiment of the present invention, the linear increase and exponential decay modes are initiated and the linear increase and until the ER marking can perform control of the cue length again. Activate exponential decay mode.

In the following “(6) Simulation Results” chapter, simulations show that the RR marking effectively limits the worst queue length in the transient period, while at the same time the ER marking re-executes the actual control to return the queue length behavior to the vibration mode. The transition to asymptotic stable mode at will be shown.

(6) Simulation results

In the following, simulation results are presented to confirm the above-described analyzes and to test the superior performance of the algorithm according to the embodiment of the present invention. The simulation model used was developed on the NIST ATM simulator platform.

We considered two different network topologies, a fair standard: peer-to-peer configuration and parking lot configuration. Table 1 below summarizes the recommended values for the design variables of the algorithm proposed by the present invention, which are used in the following simulation studies.

ER Allocation Algorithm AB q _T T Estimation Algorithm W δ λ 250 cells 0.9 0.98

( = max { , }, = Transmission time of one cell)

First we consider a peer-to-peer configuration. 10 shows a peer-to-peer model configuration including a VBR source. In Fig. 10, s1, .., s20, and vbr1 are sources for transmitting data, and d1, .., d20, and dvbr1 are destinations for receiving the transmitted data. Among the sources, vbr1 is the VBR source, and among the destinations, dvbr1 is the VBR destination.

Referring to the peer-to-peer configuration of FIG. 10, 20 ABR VCs have the same path, and each of the links has the same capacity as 600 Mbps. In addition, in order to implement a WAN environment, the distance between each of the sources s1, .., s20 of FIG. 10 and the first switch SW1 is set to 1,000 km. If we assume that the signal propagation speed is 2.5 × 10 ⁵ Km / sec and we ignore the queuing time, Is approximately 10 msec. The VC models used in the simulation configuration of FIG. 10 are summarized in Table 2 below and all sources are persistent. In Table 2 below, units of PCR, MCR, ICR, arrival time, and departure time are Mbps.

Source PCR MCR ICR Arrival Time Departure Time Equity Ratio 0 to 1 1 to 2 2 to 3 3 to ∞ s1-s4 150 0 10 0 ∞ 36.1 33.3 30 32.5 s5-s9 150 10 10 0 ∞ 46.1 43.3 40 42.5 s10 150 0 10 2 ∞ 30 32.5 s11-s14 25 0 25 0 ∞ 25 25 25 25 s15-s19 25 10 25 0 ∞ 25 25 25 25 s20 25 0 25 1 3 25 25

It should be noted that the PCR values, MCR values, ICR values, and the arrival and departure times of VCs are examined to investigate the impact of PCR-restricted sources, differences in MCR and ICR, and call activity in network performance. For comparison purposes, we calculate theoretical fairness ratios that meet the MAX-MIN fairness as well as MCR guarantees and include the results in Table 2. Referring to Table 2, it can be seen that the fair rate of each VC is changing at that time according to arrivals and departures of other VCs. Sources s1-s10 are bottlenecked in switch SW1, and sources s11-s20 are bottlenecked by their PCR restrictions. For example, the ABR source s1-s4 is 36.1 Mbps in 0 to 1 seconds, 33.3 Mbps in 1 to 2 seconds, 30 Mbps in 2 to 3 seconds, 3 seconds to 3 seconds for maximum-minimum fairness. Should transmit data at 32.5Mbps. In the present invention, 32 data cells are generated between two adjacent omni RM cells. That is, NRM = 32.

11A-11D show experimental results in a peer-to-peer model configuration with only ER marking and no VBR background traffic, and FIGS. 12A-12D show a peer-to-peer model configuration with ER marking and VBR background traffic. Experimental results are shown.

FIG. 11A is a source transfer rate a _i (t) of VC with a PCR of 150 Mbps, and FIG. 11B is a source transfer rate a _i (t) of VC with a PCR of 25 Mbps, and the actual source rates shown in FIGS. 11A and 11B are tables It can be seen that it is in full agreement with the theoretical fairness ratio in 2. The transmission rates of sources s1-s4, s10 correspond to the common ER r (t) calculated at switch SW1 because their MCR is 0Mbps, and the transmission rates of sources s5-s9 are 10Mbps higher than common ER because their MCR is 10Mbps. . And sources s11-s20 are independent of their MCR values and are limited by their PCR. The initial transient characteristic is due to the initial condition of ER value r (0) = 0 in both switches SW1 and SW2. In other words, it takes time for the common ER value to jump to the operating point, but this is rarely occurred during normal operation. In FIG. 11C, the queue length at the bottleneck node SW1 is shown. The joining of source s20 in one second and source s10 in two seconds results in a surge in cue length and a residual of source 20 in three seconds results in a sudden drop in queue length. Brings about. However, according to an embodiment of the present invention, the proposed algorithm is quickly restable at the queue length q _T = 250 cells. In FIG. 11D, the estimated number of local bottlenecks VC in switch SW1 | Q | _avg (t) Estimation By computing the smallest integer estimate greater than or equal to it by using Equation 38, we obtain the integer estimate | Q | as The symbol "[x""means" x = "and" y) "means"y>". The integer estimates are 18 in [0,1) seconds, 9 in [1,2) seconds, and [2, Is 18 in seconds. Note that except for the initial time interval [0,1) seconds, the integer estimate is 9, [2, 2] at [0,2) seconds in Table 2. | Q |, represented by 10 in seconds It is a perfect match with the truth value. The difference in time interval [0,1) seconds is estimated by 1 for every arrival of a new VC, as shown in equation (42). _This is due to the intentional increase of _avg (t). In a given simulation scenario, 18 VCs are reached at 0 seconds, which is estimated | Q | This is because _avg (t) initially has 18.

For VBR background traffic the result does not change vertically as shown in FIGS. 12A-12D. That is, macroscopic (time-average) operating characteristics are almost the same as those with non-VBR traffic disturbances. VBR background traffic has peak rates, on and off cycle lengths of 10 Mbps and 20, respectively. Seconds and 20 Generated by the determined on / off source, which is seconds. 12A and 12B, the transmission rate of the sources (except the PCR limiting source) almost completely follows the on / off pattern of the VBR background traffic. On / off behavior of VBR background traffic causes repeated surge and drop of ABR queue length. However, the ER allocation algorithm according to an embodiment of the present invention allows the queue length to quickly recover to the target value as shown in FIG. 12C.

Next, considering the parking lot configuration shown in FIG. 13, the case of VCs having different round trip delays and a plurality of bottleneck nodes is as follows. 16 ABR VCs with different source locations are included, and the performance of the links is set equal to 600 Mbps, except that the link between switches SW3 and SW4 and the link between SW4 and SW5 is 300 Mbps. The VC models used in this simulation setup can be summarized in Table 3. Here all the sources assume the persistence.

Source PCR MCR ICR Arrival Time Departure Time Equity rate Bottle Neck Location s1, s5, s9 150 0 10 0 ∞ 21.67 SW3 s13 150 0 10 0 ∞ 120 SW4 s2, s6, s10 150 10 10 0 ∞ 31.67 SW3 s14 25 0 25 0 ∞ 130 SW4 s3, s7, s11 25 10 25 0 ∞ 21.67 SW3 s15 25 0 25 0 ∞ 25 PCR s4, s8, s12, s16 25 10 25 0 ∞ 25 PCR

For purposes of comparison, the present invention calculated the maximum-minimum fairness theoretical fairness that MCR guarantees for a given scenario and the results are included in Table 3. And to clarify this scenario, we included the theoretical bottleneck locations for each VC in Table 3. The bottleneck position is at a position where each individual fair share is determined. 14A to 14F show simulation results when there is no VBR background traffic and only ER marking. 14A to 14F are experimental results in a parking lot model configuration with only ER marking and no VBR background traffic. FIG. 14A is a source transfer rate a _i (t) of VC with 150 Mbps of PCR, and FIG. 14B is 25 Mbps with PCR. Source transfer rate a _i (t) of VC, FIG. 14C is the queue length at switch SW3, FIG. 14D is the queue length at switch SW4, and FIG. 14E is an estimate of the number of local bottlenecks VC at switch SW3 | Q | _avg (t), and FIG. 14F shows the estimated number of local bottlenecks VC in switch SW4 | Q | _avg (t). Referring to Figures 14A and 14B, the actual source rates in steady state are in full agreement with the theoretical fairness given in Table 3, regardless of their round trip delays and bottleneck locations. The initial transient is due to an initial condition of r (0) = 0 in all switches, which rarely occurs during normal operation. In the given scenario, there are two congested nodes SW3 and SW4. The expected queue lengths at these congested nodes SW3 and SW4 converge to the target value 250 cells, as shown in Figs. 14C and 14D. 14E and 14F show the estimated number of local bottlenecks VC in switches SW3 and SW4, respectively. _avg (t) Our observations indicate that the estimates for SW3 and SW4 in the steady state remain around 9 and 2, respectively. This is consistent with the data in Table 3.

Lastly, when RR marking is applied together with ER marking, the results according to the embodiment of the present invention will be described. We set the queue length thresholds for RR marking to q _T = 250 cells, q _LT = 750 cells, q _HT = 1000 cells. This selection of queue length thresholds is made in anticipation that ER marking plays an important role in ensuring asymptotic normal maximum-minimum flow control, whereas RR marking plays an additional role in limiting transient overshooting of queue length. will be. An embodiment of the present invention uses the same parking lot configuration scenario with VBR background traffic. The VBR background traffic has peak rates, lengths of on and off periods of 10 Mbps and 20, respectively. Seconds and 20 Generated by the determined on / off source, which is seconds. The results are shown in FIGS. 15A-15F. 15A to 15F are experimental results of a parking lot model configuration with ER marking and VBR background traffic. FIG. 15A is a source transmission rate a _i (t) of VC having a 150 Mbps PCR, and FIG. 15B is a VC having a 25 Mbps PCR. Source transfer rate a _i (t), FIG. 15C is the queue length at switch SW3, FIG. 15D is the queue length at switch SW4, and FIG. 15E is the estimated number of local bottlenecks VC at switch SW3 | Q | _avg (t), and FIG. 15F shows the estimated number of local bottlenecks VC in switch SW4 | Q | _avg (t) As compared with the case of Figs. 14A to 14F, the transient overshoot of the ABR queue length in SW3 actually dropped from 6000 cells to 3200 cells with the maximum reduction. Clearly this drop is due to the activation of additional RR markings. However, this gain pays for the temporary vibration of queue length and source rate, as seen in FIGS. 15A, 15B and 15C. We observed that the ER assignment algorithm according to an embodiment of the present invention can be effectively controlled, and as a result, the queue length behavior is recovered from oscillatory mode to asymptotic mode.

In the above description of the present invention, specific embodiments have been described, but various modifications may be made without departing from the scope of the present invention. Therefore, the scope of the present invention should not be defined by the described embodiments, but should be determined by the equivalent of claims and claims.

As described above, the present invention guarantees maximum link utilization and minimum cell loss irrespective of the round trip time delay of the ABR loop, and minimizes the requirement of ABR queue size by ensuring asymptotic stability of the ABR queue. In addition, the present invention ensures fairness of bandwidth utilization among ABR users, and in particular, guarantees maximum and minimum fairness, which is an ATM forum standard, and can quickly adapt to changes in the network environment such as changes in the number of ABR users and changes in ABR bandwidth. In addition, it performs almost all the functions suggested by the ATM Forum traffic management standard, including functions such as EFCI marking, RR marking, and ER marking, and has an asymptotic stabilizing operating point to enable high availability, low cell loss, and maximum-minimum fairness. Allows (MAX-MIN fair rate) allocation to be achieved. In addition, the present invention improves responsiveness and control transient performance and calculates algorithms for multiple time scales, i.e., changes in cell level rates of VBR and ABR VC and cell load arrival and departure of VBR and ABR VC. There is an advantage that the implementation complexity and scalability is minimized to minimize the number of operations required to eliminate and eliminate VC operations including VC queuing, VC calculations, and VC table accesses.

Claims (1)

The packet switched network includes a plurality of nodes connected to each other, each node is connected to a plurality of sources for transmitting and receiving data, and has a data queue for storing data transmitted from the source, the data queue is the current queue length And a method for fair flow control in the packet switched network, related to a target value, Estimating the number of local bottleneck virtual circuits by each node using the data rate in the corresponding source and the minimum data rate to be guaranteed; Assign a data rate to each of the sources based on a difference between the current queue length and a target value, the differential value of the current queue length, and the estimated number of local bottleneck virtual lines, wherein the data rate is each of the sources. Fair flow control method, characterized in that consisting of a process that is assigned through a feedback signal transmitted to.