WO1996024212A1

WO1996024212A1 - Bandwidth management and access control for an atm network

Info

Publication number: WO1996024212A1
Application number: PCT/US1996/001208
Authority: WO
Inventors: Wei Chen
Original assignee: Bell Communications Research, Inc.
Priority date: 1995-02-03
Filing date: 1996-01-23
Publication date: 1996-08-08
Also published as: CA2186449A1; EP0754383A1; CA2186449C; KR970702642A; KR100222743B1; US5533009A; TW344920B; JPH09505720A; JP3088464B2; EP0754383A4; MY112027A; EP0754383B1

Abstract

A bandwidth management system (120) manages a plurality of virtual data connections within a communications network. The system includes an input (122) for receiving data cells, wherein each cell is associated with a particular one of the virtual connections. The system also includes a cell pool, coupled to the input for storing the cells, first and second queues (126-128) for ordering the virtual connections, and an output (137) for transmitting cells from the cell pool. The relative position of a virtual connection in the first queue is determined by an eligibility variable that varies according to an anticipated data rate associated with the particular virtual connection and according to an amount of time that the particular virtual connection has been in the first queue. The relative position of a virtual connection in the second queue varies according to a predetermined quality of service that is assigned to each of the virtual connections.

Description

BANDWIDTH MANAGEMENTANDACCESS CONTROL FORANATM NETWORK

Government Rights

This invention was made with Government support under Contract DABT63-93-C-0013 awarded by the Department of the Army. The Government has certain rights in the invention.

Field of the Invention

This invention relates to the field of asynchronous data communication and more particularly to the field of managing the flow rate of data through an asynchronous data communication network.

Related Art

The next generation public backbone network capable of supporting voice, video, image, data, and multi-media services is envisioned as a broadband Integrated Services Digital Network (ISDN) using an asynchronous transfer mode (ATM) to transmit data. The perceived advantages of ATM technology include flexibility and simplicity of asynchronously multiplexing traffic sources with a very broad range of source parameters and service quality requirements with information loss and delay performance ranging from close to that of the synchronous transfer mode to that of best-effort service in today's packet-switched networks. Broadband ISDN is therefore a promising network technology to support emerging needs of high performance computing and communications. The major technical impediment to practical large scale deployment of ATM-based broadband ISDN networks, however, is the lack of practical real-time bandwidth management that will guarantee predictable end-to-end grades of service over a multi-node, multi-carrier network to end application platforms, such as high performance workstations and supercomputers.

In ATM-based broadband ISDN, information from the application layer is processed at the ATM Adaptation layer into fixed size ATM cells, which are in turn multiplexed/switched at the ATM layer, and transported in payload envelopes at the physical layer. Performance degradation caused by congestion due to insufficient network resources at certain parts of the network leads to perceptible performance degradation to the end-users, unless robust bandwidth management policies are defined and implemented. The overall objective of bandwidth management and traffic control of broadband ISDN is to provide consistent and predictable end-to-end performance, while at the same time optimizing network resource utilization for end-users with a wide range of performance requirements and traffic flows, including some which are very bursty.

The problem of bandwidth management and traffic control of broadband ISDN becomes complex due to factors such as the diversity of traffic flow characteristics (much of which is unknown even at the time of transmission), connection performance requirements, the impacts of end-to-end propagation delay and the processing delay at the network elements due to increased switching and transmission speeds. The very high speed of ATM networks, coupled with propagation delay results in large numbers of ATM cells in route, while the (relatively) large cell processing time at intermediate nodes imposes severe limitations on intermediate node protocol processing. These factors cause existing traffic control approaches (like the X.25 data communications flow control designed for traditionally low-speed networks) to become ineffective. Hence new, robust (to cope with unknown and changing traffic characteristics), low-cost and scalable (to increasingly higher speeds) traffic control strategies are needed for Broadband ISDN.

The real-time (typically several milliseconds or less) traffic control requirements of broadband ISDN is drastically different from those of existing networks due to the factors discussed above. According to functionality, the real¬ time traffic control may consist of well-coordinated components such as access control, flow control, reactive congestion control, and error control. Access control, including connection admission control/bandwidth allocation, traffic shaping, and bandwidth enforcement, can achieve user objectives at the user-network interface. Other controls (flow control, reactive congestion control, error control) achieve destination buffer protection, network internal congestion handling, and information integrity guarantee, respectively.

The central goal of access control is to achieve, at the user-network interface, objectives such as predictable information throughput, connection blocking probability, cell loss ratio, and cell transfer delay/cell delay variation, among others. In view of uncertain and changing traffic characteristics/parameters and increased transmission/switching speeds in broadband ISDN, the access control mechanism should be robust, fast, scalable (from 155 to 622 Mbps, or higher), and low-cost to implement.

There are three related aspects of access control: connection admission control, bandwidth shaping and pacing, and bandwidth enforcement. Connection admission control decides connection acceptance, given traffic characteristics and performance requirements (e.g. peak rate, cell loss rate), based on the current network state. Generally, a decision needs to be made and network resources allocated in no more than a few seconds. Some key issues include what traffic descriptors should be used (i.e., peak data rate, average data rate, priority, burst length, etc), how to determine effective bandwidth of a bursty connection, how to predict performance (given underlying traffic shaping/bandwidth enforcement mechanisms), what acceptance/rejection criteria to use, and how fast can algorithms be executed.

Currently, different solutions for connection admission control exist, ranging from peak rate allocation (simple to implement, but bandwidth inefficient for variable bit-rate services) to more sophisticated algorithms which determine the "admissible acceptance region" in an N-service space. Typically, the connection admission control will be implemented in software, which makes it possible to start with a simple and robust algorithm during the early stages of broadband ISDN and yet allows evolution towards more sophisticated (and effective) algorithms as they become available. A distinguishing feature of access control schemes is their capability to provide statistical multiplexing for variable bit rate services. A number of factors significantly impact the statistical multiplexing efficiency including, among others, variable bit-rate transmission peak rate (relative to the link capacity) and the burst duration distribution. Studies have shown that pure statistical multiplexing of variable bit-rate connections may result in a low (e.g. 20-25%) network utilization, if no traffic shaping or resource allocation (such as the fast reservation protocol) techniques are used. By proper modification of the cell arrival process (i.e., traffic shaping), higher statistical multiplexing efficiency may be achieved. Such modification techniques employed by the pacing mechanism may include cell jitter smoothing, peak cell rate reduction, message burst length limiting, and service scheduling of variable bit rate virtual circuit connections (VCs), among others.

Further, traffic shaping may possibly be used in conjunction with a network bandwidth enforcement mechanism by rescheduling a cell's service (in addition to cell discarding or tagging) when a non-compliant cell is observed. For non-delay sensitive but loss sensitive variable bit rate services, this option is appealing in order to minimize the cell loss ratio. However, applicability of pacing techniques to cells with stringent delay requirement (such as interactive variable bit rate video) requires further study. From a network operation point of view, source shaping/pacing is also desirable to prevent overwhelming the system with more data than the system can handle.

The purpose of network bandwidth enforcement is to monitor a connection's bandwidth usage for compliance with appropriate bandwidth limits and to impose a policing action on observed violations of those limits. The enforcement control works on the time scale of a cell emission time, (i.e., about 2.7 μsec for 155 Mb/s service or about 0.7 μsec for 622 MB/s service). Key issues of bandwidth enforcement include the design of the monitoring algorithm and policing actions to be taken on non-compliant cells. Other issues include handling of traffic parameter value uncertainty, the effectiveness in terms of the percentage of erroneous police actions upon compliant cells and the percentage of non-compliant cells undetected, and the detection time of a given violation.

The bandwidth enforcement mechanism operates upon network-measured traffic parameter values, which may include a connection's peak cell rate, its average cell rate, and its peak burst length. Currently, a few bandwidth enforcement algorithms have been proposed using, for example, single or dual leaky buckets, jumping, or sliding windows. However, some studies have shown that the leaky-bucket algorithms, using peak/average rates and average burst length, may still not be robust enough for various variable bit rate traffic mixes. The studies also show that the leaky-bucket algorithms also tend to complicate performance prediction at the connection admission control level. Furthermore, accurate estimation of average cell rate and burst length by users may be very difficult in practice. This suggests the need for the exploration of alternative approaches to this problem, in order to bring about an early deployment of broadband ISDN. A number of access control approaches have been proposed, but there is no standard consensus among the vendor and research communities. An approach has been proposed that accepts connections based on bandwidth pools dedicated to several traffic classes respectively and uses a leaky bucket type algorithm for monitoring connection bandwidth utilization with immediate cell discarding of non-compliant cells. Another proposal also uses a leaky bucket type monitoring algorithm with tagging of non-compliant cells for possible later cell discarding. Another proposed approach uses a rate-based time window approach.

One study has indicated that the leaky bucket algorithm is superior to the time-window based approaches under certain traffic patterns studies. However, the same study also revealed difficultly in its dimensioning (e.g., the counter limit). Further, the performance of the leaky bucket algorithm is also found to be far below optimal, in terms of non-compliant cell detection and false alarms (the long term probability of declaring a bandwidth violation for compliant cells).

This seems to indicate that enforcement and pacing near the average cell rate for variable bit-rate services are much more complex than that has been generally recognized. For example, enforcement of a variable bit-rate traffic stream near the average rate generally leads to enormous buffer requirements (hence unacceptable response time) in order to keep the false alarm rate to an acceptable level (say 10^"7).

More importantly, the effectiveness of a monitoring algorithm (such as the leaky bucket) is critically dependent on the source model behavior which is unknown for many applications and is difficult to estimate accurately at the time that the connection is made. Other reservation based schemes are more suitable for data traffic, but their performance characteristics have not been fully evaluated.

SUMMARY OF THE INVENTION

According to the present invention, a bandwidth management system manages a plurality of virtual data connections within a communications network. The system includes an input for receiving data cells, wherein each cell is associated with a particular one of the virtual connections. The system also includes a cell pool, coupled to the input, for storing the cells, a first and second queue for ordering the virtual connections, and an output for transmitting cells from the cell pool. The relative position of a virtual connection in the first queue is determined by an eligibility variable that varies according to an anticipated data rate associated with the particular virtual connection and according to an amount of time that the particular virtual connection has been in the first queue. The relative position of a virtual connection in the second queue varies according to a predetermined quality of service that is assigned to each of the virtual connections. The output transmits a cell from the cell pool corresponding to a virtual connection at the front of the second queue.

Virtual connections having eligibility variables with equal values can be ordered in the first queue according to the predetermined quality of service. The system can use four different values for quality of service. Virtual connections having equal quality of service values can be ordered in the second queue according to a priority computed for each of the virtual connections. Credit variables for each virtual connection can indicate allocated time slots provided to each of the virtual connections. The priority can vary according to one or more of: one or more anticipated data rates of each of the virtual connections, the value of one or more of the credit variables, and the number of bac logged cells that are awaiting transmission. The data rates, credit variables, and backlog values can be weighted prior to determining the priority. The system can use a burst bit indicator to determine if virtual connections associated with cells received at the input should

be placed immediately on the second queue.

The bandwidth management system can be one of: a pacing unit and an enforcement unit. A pacing unit receives cells from a data source node and provides cells to a communication link. An enforcement unit receives data from a communication link and provides data to a data sink node. A virtual connection

can be established by specifying initially agreed-upon traffic parameters and dropping cells that are received at a rate that exceeds that specified by the initial

traffic parameters.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention is better understood by reading the following Detailed Description of the Preferred Embodiments with reference to the accompanying drawing figures, in which like reference numerals refer to like elements throughout, and in which:

FIG. 1 shows a data communications network having a plurality of physically interconnected nodes.

FIG. 2 shows a plurality of communication nodes interconnected by a plurality of virtual connections.

FIG. 3 is a data flow diagram showing different states of data within a bandwidth management unit according to the present invention.

FIG. 4 is a functional block diagram of a bandwidth management unit according to the present invention.

FIG. 5 is a state diagram showing operation of a bandwidth management unit according to the present invention.

FIG. 6 is a state diagram illustrating setting up a virtual connection. FIG. 7 is a flow diagram illustrating cell admission for a virtual connection.

FIG. 8 is a flow diagram illustrating initializing the value of an eligibility timer for a virtual connection.

FIG. 9 is a flow diagram illustrating resetting the value of an eligibility timer for a virtual connection.

FIG. 10 is a flow diagram illustrating updating credit variables of a virtual connection.

FIG. 11 is a flow diagram illustrating updating an eligibility timer of a virtual connection.

FIG. 12 is a flow diagram illustrating computing values for credit variables and a priority variable for a virtual connection.

FIG. 13 is a flow diagram illustrating updating a priority variable for virtual

connection.

FIG. 14 is a schematic diagram illustrating hardware implementation of a bandwidth management unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In describing preferred embodiments of the present invention illustrated in the drawings, specific terminology is employed for the sake of clarity. However, the invention is not intended to be limited to the specific terminology so selected, and it is to be understood that each specific element includes all technical equivalents which operate in a similar manner to accomplish a similar purpose.

Referring to FIG. 1, a communications network 30 comprises a plurality of communication nodes 32-50 that are physically interconnected. Each of the nodes 32-50 represents conventional communications equipment for sending and receiving communications data. The lines drawn between the nodes 32-50 represent physical communications links for transmitting data between the nodes 32-50. The communications links can be any one of a plurality of conventional mediums for communication data transmission including fiber optics, twisted pair, microwaves links, etc.

It is possible for two nodes to be directly connected to each other, which facilitates communication between the directly connected nodes. For example, the node 34 is shown in FIG. 1 as being directly connected to the node 35. Accordingly, communications between the node 34 and the node 35 is facilitated via the direct physical communication link.

Often, it is desirable to communicate between two nodes which are not directly connected via a single physical communication link. ^"For example, it may be desirable to send data from the node 34 to the node 49 of FIG. 1. In that case, communication is facilitated by passing messages through intermediate nodes between the node 34 and the node 49. Data from the node 34 to the node 49 can be sent by first transmitting the data from the node 34 to the node 35, then from the node 35 to the node 36, then from the node 36 to the node 45, and finally from the node 45 to the node 49.

In the example above, a "virtual connection" between the node 34 and the node 49 is established to facilitate communication between the nodes 34, 49 and to establish desired quality of service for the communication. A virtual connection is established between two nodes when one of the nodes has data that is to be sent to another node. A virtual connection can be terminated after all of the data has been sent. For example, if the node 34 represents a user computer terminal and the node 49 represents a mainframe computer, then a virtual connection can be established between the node 34 (terminal) and the node 49 (mainframe) whenever a user logs on to the computer system. The virtual connection could then be terminated when the user logs off the system.

Referring to FIG. 2, a schematic diagram 60 shows a plurality of hosts 62- 67, a pair of private ATM (asynchronous transfer mode) mux/switches 71, 72, and a public ATM mux/switch 74. Each of the hosts 62-67 represents one of a plurality of possible sources of data, such as a telephone, a fax machine, a computer terminal, a video teleconference terminal, etc. The host 62 transmits data to the private ATM mux/switch 71 via a virtual connection set up between the host 62 and the private ATM mux switch 71. The host 63 transmits data to d e mux/switch 71 via a second virtual connection between the host 63 and the private ATM mux/switch 71. A virtual connection can span a single direct physical link or can span a plurality of consecutive physical connections through intermediate network nodes, not shown in FIG. 2. Similarly, the host 64, 65 are connected to the mux/switch 72 via separate virtual connections therebetween. The mux switches 71, 72 and the hosts 66, 67 are connected to the public ATM mux/switch 74 via additional virtual connections. The public ATM mux/switch 74 provides data to a trunk (not shown).

The diagram 60 shows a general flow of data from the hosts 62-67 to the trunk via the intermediate devices 71, 72, 74. The virtual connections shown in the diagram 60 are set up at the time that the hosts 62-67 begin transmitting data. For example, if the host 62 represents a telephone, then a virtual connection between the host 62 and the private ATM mux/switch 71 can be established by a connection admission procedure when the user picks up die handset of the telephone and begins dialing. The virtual connection can be terminated when the user hangs up the telephone.

In order to avoid exceeding the data-handling capacity of either the trunk, the switches 71, 72, 74 or the links tfierebetween, it is desirable to control the amount of data that flows from the hosts 62-67 to the switches 71, 72, 74 and to control the amount of data that flows from the switches 71, 72, 74 to the trunk. Accordingly, each output port of the hosts 62-67 is provided with one of a plurality of pacing units 82a-87a. The pacing unit 82a is connected to the output port of the host 62, the pacing unit 83a is connected to the output port of the host 63, etc.

The pacing units 82a-87a control the data transmitted from the hosts 62-67 by limiting all data transmitted on the associated connection. For example, the pacing unit 82a limits data transmitted from the host 62 to the private ATM mux/switch 71. Data is first passed from the host 62 to the pacing unit 82a via the output port of the host 62. The pacing unit 82a then provides the data to the communication link.

The pacing unit 82a provides the data on the communication link at an agreed upon rate, as described in more detail hereinafter. Note that the switches

71, 72 also transmit data via pacing units 91a, 92a connected to the output ports thereof.

Associated with each of the pacing units 82a-87a, 91a, 92a are corresponding enforcement units 82b-87b, 91b, 92b. The enforcement units 82b- 87b, 91b, 92b are at the receiving ends of the communication links to ensure that the agreed upon data rate of each of the connections is maintained. The enforcement units 82b-87b, 91b, 92b are connected to input ports of corresponding devices so that, for example, the enforcement unit 82b is connected to the input port of the mux/switch 71. If a pacing unit provides data at greater ti an the agreed upon data rate, the corresponding enforcement unit discards the excess data.

Pacing units are provided at data source nodes while enforcement units are provided at data sink (i.e., receiving) nodes.

When a virtual connection is established between two nodes, the pacing unit at the data source node and the enforcement unit at the data sink node initially agree upon a data rate and a quality of service. Quality of service relates to a variety of characteristics, including the acceptable delay between generation and reception of the data, as described in more detail hereinafter. The requested data rate and quality of service are functions of the type of data being sent. For example, real time voice data may require a different data rate than would real time video data, but both may require a relatively high quality of service. In

contrast, an E-Mail message would probably require a relatively low quality of service.

During initialization of the virtual connection, a pacing unit requests an appropriate set of traffic parameters (such as data rate, buffer space, etc.) and requests the desired quality of service. The corresponding enforcement unit at the other end of the virtual connection can accept or deny the request of the pacing unit A request is denied if the node to which the enforcement unit is attached is not capable of handling the requested amount of data to meet the requested quality of service. Reasons for denying the request include insufficient capacity of the data sink node to handle the requested data rate and insufficient capacity of the communications link to handle the requested data rate. If the request is denied by the enforcement unit, then the virtual connection is not established. Assuming however that the request is accepted, die virtual connection between the nodes is established. Once established, the pacing unit operates so as to send data according to the initially agreed upon traffic parameters.

Data is communicated over the communications links via packets or fixed- size ATM cells. Each cell contains one or more bytes of digital data along with header information identifying the virtual connection and other appropriate information, discussed in detail hereinafter. Note u at, for d e network shown in FIG. 2, data rate limiting all of die connections between the nodes will, of necessity, limit the total amount of data diat is provided to the trunk. Accordingly, the pacing units 82a-87a, 91a, 92a and enforcement units 82b-87b, 91b, 92b can be used to prevent the trunk, and hence the network, from being overloaded with too much data.

The pacing and enforcement units receive cells at their respective inputs and transmit cells from their respective outputs. The operation of each is substantially identical. Accordingly, the detailed discussion which follows is equally applicable both types of units which are referred to generically as bandwidth management

units.

Referring to FIG. 3, a data flow diagram 100 illustrates different states of a virtual connection handled by a bandwidth management unit. Note that a single bandwidth management unit can handle multiple virtual connections even though the bandwidth management unit is physically connected to only a single transmission link. The node to which die bandwidth management unit is connected sends or receives cells for the various connections and the bandwidth management unit outputs the cells in a particular order.

Initially, a cell is provided to the bandwidth management unit from either the output port of the associated data source node (for a pacing unit) or from the communication link (for an enforcement unit). The associated virtual connection

(VC) begins in an unscheduled state 102. A cell corresponding to a VC in the unscheduled state 102 is deemed not to be ready to transmit by the bandwidth management unit.

The VC transitions out of the unscheduled state 102 by becoming eligible and having at least one data cell ready to transmit. Eligibility of a VC is controlled externally by the data node connected to the bandwiddi management unit in a manner described in more detail hereinafter. A VC which is deemed immediately eligible by the data node will transfer out of the unscheduled state 102.

The VC can transition from the unscheduled state 102 to an eligibility state

104. The eligibility state 104 indicates that die cells associated with die VC are waiting to become eligible for output, but the VC is waiting for particular conditions to be met, which are described in more detail hereinafter.

When the conditions are met, the VC can transition from the eligibility state 104 to a departure state 106. A VC in the departure state 106 has at least one cell associated therewith which is ready to be output by the bandwidth management unit. The cells of the VC in the departure state are transmitted by the bandwidth management unit in a specific order, as described in more detail hereinafter.

A VC which is in the eligibility state 104 or in the departure state 106 can be transitioned back to the unscheduled state 102 when the data node connected to the bandwiddi management unit deems the VC to be ineligible or when other conditions, described in more detail hereinafter, are met. Furthermore, it is possible for a VC in the departure state 106 to be transitioned back to the eligibility state 104 under certain conditions, also described in detail hereinafter. Under certain odier conditions, a VC in the unscheduled state 102 can transition directly to die departure state 106 witiiout going through the eligibility state 104. Transitions between various states and the conditions therefor are described in more detail hereinafter.

The ordering of the VC's in a queue associated witii die departure state 106 is a function of the Quality of Service (QOS) of die virtual connection. The QOS of a virtual connection is established when the virtual connection is initialized by die data source node. The source node for die data can request one of four possible values for the QOS: high level QOS, mid level QOS, low level QOS, and best effort QOS. High level QOS is for real time traffic that requires bounded delay and virtually no cell loss. Mid level QOS is for real time traffic that requires bounded delay and little cell loss. Low level QOS is for delay- tolerant data traffic having controllable cell loss. Best effort QOS is for data which can be transmitted at a very low rate that can share available bandwiddi not being used by die high level, mid level, and low level QOS cells. The ordering of the cells in the queue is based on the QOS associated with the virtual connection and upon other parameters which are discussed in detail hereinafter. Note diat, altiiough the invention is illustrated herein witii four separate levels for die QOS, the invention can be practiced with a different number of levels. The number of levels chosen for the implementation is a design choice based on a variety of functional factors known to one of ordinary skill in die art.

Referring to FIG. 4, a schematic diagram 120 functionally illustrates operation of a bandwidth management unit. An input 122 accepts cells from a data node (not shown) connected to die bandwiddi management unit. The cells diat are received are provided to a demultiplexer 124. The demultiplexer 124 separates die input cells according to die VC associated tiierewith and places the cells in one of a plurality of appropriate queues 126- 128, so diat, for example, cells associated with virtual connection VC, are placed in the queue 126, cells associated with virtual connection VC_j are placed in die queue 127 and cells associated the virtual connection VC_k are placed in die queue 128. A cell admission unit 130 controls placing d e cells in the queues 126-128 in a manner described in more detail hereinafter.

Associated witii each virtual connection, V , VC_r and VC_k, is a set of variables 131-133 diat controls when the cells for each of die VC's will be transmitted out of die bandwiddi management unit. Also when a cell is input to die bandwiddi management unit, the cell admission unit 130 examines the variables 131-133, to determine if me cell should be admitted. As shown in FIG. 4, the virtual connection V has the variable set 131 associated tiierewitii, the virtual connection VC_j has die variable set 132 associated tiierewitii, and the virtual connection VC_k has die variable set 133 associated tiierewitii. The variable sets

131-133 are described in more detail here and after.

The variable sets 131-133 are provided as inputs to a mux control 135 which controls which cell from the head of one of die queues 126-128 is next to be transmitted out of d e bandwiddi management unit. A multiplexor 136 is connected to each of die queues 126-128 and, in conjunction witii die mux control

135, provides a cell from one of die queues 126-128 to an output 137 for transmission out of the bandwidth management unit.

The table below facilitates die discussion which follows by showing parameters diat are used for each virtual connection of the bandwiddi management unit:

VARIABLE NAME DESCRIPTION scd, state of VC, qos, quality of service for VC, e, indicates if transmission for VC, is enabled s, eligibility timer for VC, r,l first rate variable for VC, r,2 second rate variable for VC, c,l first credit variable for VC, c,2 second credit variable for VC,

1, number of backlogged cells for VC .l first weight variable for VC, w,2 second weight variable for VC, w,3 third weight variable for VC, w,4 fourth weight variable for VC, w,5 fifth weight variable for VC, b, burst bit indicator for VC,

P, priority variable for VC,

Referring to FIG. 5, a state diagram 140 illustrates operation of a bandwiddi management unit. Although operation is illustrated witii regard to a single VC, VC,, die operation of other VC's of die bandwiddi management unit is identical.

Circles on die diagram 140 show the tiiree possible states of VC,: die unscheduled state, die eligibility state, and die departure state. A VC is always in one of those tiiree states, although for a bandwidth management unit servicing multiple VC's, die state of any one of die VC's is independent of the state of any of the odier VC's. The state diagram 140 also shows rectangles having rounded corners which represent operations performed by die bandwiddi management unit with respect to VC,. Diamonds on the diagram show decision branches where die flow is conditioned upon a particular variable or variables associated witii VC, being in a particular state or having a particular value or values. The arrows between the various circles, rounded rectangles, and diamonds illustrate the flow through the state diagram. For an arrow having an annotation associated tiierewith, the annotation indicates die condition causing the transition from one portion of the state diagram 140 to another. Arrows having no annotation associated therewith represent an unconditional flow from one portion of die state diagram 140 to anodier.

Flow begins at an initial step 142 where VC, is initialized. Initialization of a VC includes having a pacing unit request a particular set of traffic parameters and quality of service (QOS) from die associated enforcement unit (i.e., the enforcement unit at die odier end of die communication link). That is, VC, is initialized when the pacing unit sends a command to die associated enforcement unit requesting creation of VC, and specifying die set of traffic parameters and

QOS for VC,. As discussed above, die enforcement unit either accepts or rejects die request depending on die existing data loading on both die communication link and die receiving node. The decision is based on die loading for die node attached to die enforcement unit. If the request from die pacing unit to the enforcement unit is denied, tiien VC, is not initialized and no processing (i.e., no data transmission) occurs for VC,.

Assuming diat die VC, initialization request is approved by die enforcement unit, then following the initialization step 142, VC, enters an unscheduled state 144. While VC, is in the unscheduled state 144, die variable scd, is set to zero. VC, remains in die unscheduled state 144 until die first cell of data for VC, is received by die bandwiddi management unit.

When a cell associated witii VC, arrives, the bandwiddi management unit transitions from the unscheduled state 144 to a cell admission step 146, where die cell is eitiier accepted or rejected by the bandwiddi management unit. The cell admission step 146 is described in more detail hereinafter.

Following die cell admission step 146 is a test step 148 which determines whedier die VC, transitions from the unscheduled state 144 to another state. At the test step 148, die variables 1, and e, are examined. The variable 1, represents die backlog, in number of cells, for VC,. That is, 1, equals die number of cells received by die bandwiddi management unit for VC, diat have not been output by the bandwiddi management unit. The variable e, is a variable indicating whedier VC, has been enabled by die data node diat is being serviced by die bandwidth management unit. That is, die data node diat is connected to die bandwiddi management unit can either enable or disable VC,. For a virtual connections that is not enabled, die bandwiddi management unit receives inputted cells but does not output the cells.

At die test step 148, it is determined if die backlog of cells for VC, is at least one (i.e., there is at least one cell associated witii VC, stored in die bandwiddi management unit) and if VC, is enabled. If one or botii of tiiose conditions are not met, then VC, returns to the unscheduled state 144.

Note also that while in the unscheduled state 144, it is possible for die bandwiddi management unit to receive a signal from die data source node requesting diat e, be set. In that case, control transfers from the unscheduled state

144 to a step 150 where e, is set. Following die step 150 is die test step 148, discussed above. Note, dierefore, that it is possible for the bandwiddi management unit to receive one or more cells for VC, prior to VC, being enabled and tiien for the bandwiddi management unit to receive an enable signal from the data node so diat VC, transitions from the unscheduled state 144.

While in die unscheduled state 144, it is possible for die bandwiddi management unit to receive a signal from the data node requesting termination of VC,. Termination can occur for a variety of reasons, including there being no more data to transmit over the virtual connection. In that case, control transfers from the unscheduled state 144 to a step 152 where VC, is terminated. Once VC, has been terminated, die bandwiddi management unit does no further processing for VC, and accepts no more cells for transmission via VC,.

It is also possible that, while in die unscheduled state 144, die bandwiddi management unit receives a request from die data node to update parameters associated with VC,. In diat case, control transfers from the step 144 to a step 154 where die parameters associated witii VC, are modified. The parameters diat can be associated witii VC, are discussed elsewhere herein. Note, however, diat one of die parameters diat can be updated at die step 154 is the quality of service associated witii the VC,, qos,. That is, die data node can request diat die quality of service associated widi VC, be changed to a second value after the VC, has been initialized widi a qos, having a first value. Following the update step 154, VC, returns to die unscheduled state 144.

If at the test step 148 it is determined diat die backlog for die VC, is at least one (i.e., 1, is greater tiian or equal to one) and diat VC, is enabled (i.e., e, equals one), then control transfers from the step 148 to a step 156 where scd, is set to one. The scd, variable associated with VC, equals zero when the VC, is in d e unscheduled state and equals one when VC, is in any of die otiier states.

Following the step 156 is a step 158 where the credit variables associated widi VC„ c,l and c,2, are updated. The credit variables, c,l and c,2, represent the amount of transmission channel bandwidth "credited" to VC,. The credit variables are described in more detail hereinafter.

Following die step 158 is a step 160 where die value of s, is initialized. The value of s, represented die amount of elapsed time that VC, is to spend in die eligibility queue before transitioning to die departure queue. That is, after exiting from the unscheduled state, VC, is placed in die eligibility queue until an amount of time indicated by s, has elapsed, after which VC, is moved from die eligibility queue to the departure queue for output by die bandwiddi management unit. The initial value of s, is determined by die associated data node and varies according to die initial traffic parameters specified when VC, is initialized.

Following die step 160 is a step 162 where die estimated values of the credits, c,l and c,2 and die estimated value of die priority, P,, associated with VC, are computed. The priority, P„ is used to order die VC's in die departure queue. Computing the priority is described in more detail hereinafter.

Following die step 162 is a test step 164 which determines if s, was set to zero at die step 160. As discussed above, a VC initializes s, according to values of the initial traffic parameters and die credits. If s, is set to zero at die step 160, tiien VC, is placed immediately on die departure queue. Otiierwise, VC, is placed in the eligibility queue for an amount of time represented by die initial setting of s, provided at d e step 160.

If at die test step 164 it is determined that s, does not equal zero, then control passes from the test step 164 to a step 166 where die value of s, is incremented by s(t). The quantity s(t) varies according to elapsed system time and is updated every cycle d at the bandwiddi management unit software is executed. In odier words, die quantity s(t) varies according to die value of a real time system clock. Since s, is set at the step 160 to represent the amount of time VC, remains on die eligibility queue, tiien adding die present value of s(t) to s, sets die value of s, to the value that s(t) will equal when VC, is to be moved from die eligibility queue to die departure queue. For example, assume mat die bandwidth management unit runs once every microsecond and diat s(t) is incremented once every cycle. Further assume d at a cell corresponding to a particular VC has been admitted and that die VC is to be moved from die eligibility queue to the departure queue in one hundred microseconds. Accordingly, die value of S_j will be set to one hundred at die step 160. At die step 166, the present value of s(t) is added to one hundred in order to provide a value of s_s diat is tested against die real time clock so as to determine when to move die VC from die eligibility queue to die departure queue.

Also at die step 166, VC, is placed in die eligibility queue. For purposes of the state diagram 140, die eligibility queue is referred to as die "A" queue. The

VC's in die eligibility queue are sorted by die value of s for each of die VC's and by the value of the quality of service for each of the VC's. In other words, the VC widi die lowest value for s is at die head of die queue, the VC with die second lowest value for s is second in die queue, etc. For VC's diat have the same value of s, the VC widi die highest QOS is first, the VC with the second highest QOS is second, and so forth. If two or more VC's have an identical value for both s and QOS, tiien the VC's are sorted in die queue in a last in first out order. The significance of die order in die cells in die eligibility queue is discussed in detail hereinafter.

Following the step 166, VC, enters die eligibility state at a step 168. Note diat scd, is set to one at die step 168. From die eligibility state 168, it is possible for VQ to transition to a step 170, which sets Q, to either one or zero according to an enable or disable command provided by die node attached to die bandwidth management unit. Note that after setting βj at the step 170, VC, returns to the eligibility state 168 without performing any further tests on die value of e_t. The value of e_t is tested at a later point in die processing (discussed below), at which time VC, is returned to die unscheduled state if e_s does not equal one. The bandwiddi management unit can receive a request to update VC,, in which case control transfers from die eligibility state 168 to a step 172 where die VQ parameters are updated. Following the step 172, die bandwidth management unit returns to die unscheduled state 168.

It is also possible for die bandwiddi management unit to receive a command, while in die eligibility state 168, to update s, for VC,. When tiiis occurs, VC, transitions to a step 174 where the value of s, is updated. Following

die step 174 is a test step 176 which determines if die update of s, has been accepted. If not, tiien control passes form the step 176 back to die eligibility state

168. Odierwise, die system transitions from the step 176 to a step 180 where the priority variable, P_it is updated. Following die step 180, VQ returns to die eligibility state 168.

Whenever VC, is in die eligibility state 168 and a new cell for VC, arrives, control transfers to a step 181 where die priority for VC,, P,, is updated.

Calculating and updating P, is described in more detail hereinafter. Every cycle that die bandwiddi management unit runs, the first N VC's at die head of die queue are tested to determine if s(t) > s,. N is a predetermined constant value, such as eight. When die value of s(t) is greater tiian or equal to the value of s_t, then the system transitions from the eligibility state 168 to a step 182 where VC, is removed from die eligibility queue (i.e., die "A" queue). Following the step 182 is a test step 184 which determines if die backlog for the VC, is at least one (i.e., 1, > one) and if VC, is enabled (i.e., e, = one). Note diat it is possible for VC, to have become disabled while in die eligibility state 168 if die bandwiddi management unit received a signal to set e, to zero at die step 170, discussed above.

If at die test step 184 it is determined diat die backlog does not contain at least one cell or diat VC, is disabled, tiien control passes from the test step 184 to a step 186 where scd, is set to zero. Following die step 186, VC, returns to die unscheduled state 144, and waits for either a cell admission or a signal to set e, to one, as described above in connection widi die description of die steps 146, 150.

If at die step 184 it is determined diat die backlog of cells is at least one and diat VC, is enabled, tiien control passes from the step 184 to a step 190 where VC, is placed in die departure queue, designated as die "B" queue on die state diagram 140. Note mat the step 190 also follows the step 164, discussed above, when it is determined at die test step 164 diat s, equals zero.

The VC's in die departure queue are sorted primarily in order of the quality of service (QOS) for die VC's. That is, VC's widi higher requested quality of service are ahead of VC's witii lower requested qualities of service. VC's in the queue having the same quality of service are sorted in order of priority, P. VC's having botii the same quality of service and die same value for priority are sorted in a first in first out basis. Following die step 190, VC, enters a departure state 192.

While in die departure state 192, VC, can receive signals to set e, and update the parameters of VC,. Setting e, is performed at a step 194 and updating the parameters for VC, is performed at a step 196. The steps 194, 196 are analogous to die steps 170, 172 discussed above in connection widi die eligibility state 168. Note that following eitiier the step 194 or the step 196, VC, returns to the departure state 192.

VC, reaches d e head of die departure queue when VC, has die highest QOS and highest priority of all of die VC's in die departure queue. In diat case, VC, transitions from the departure state 192 to a step 198 where VC, is removed from die departure queue. Following the step 198 is a step 200 where die head of die line cell (i.e., die oldest cell) corresponding to VQ is output by die bandwiddi management system. Also at die step 200, die value of die backlog, 1„ is decremented by one and the credit variables, c,l and c,2, are also decremented. Following die step 200 is a step 202 where die credit variables, c,l and c,2, are updated. Updating die credit variables is discussed in more detail hereinafter.

Following the step 202 is a step 204 where die value of s, is reset. Resetting the value of s, is discussed in more detail hereinafter. Following the step 204 is a step 206 where die variables c,l, c,2 and P,, are computed in a manner discussed in more detail hereinafter.

Following die step 206 is a test step 208 where die value of s, is examined. If die value of s, is not zero, tiien control transfers from die test step 208 to the step 166, discussed above, where the value of s(t) is added to s, and where VC, is added to die eligibility queue.

If at the step 208 the value of s, is determined to be zero, tiien control passes from the step 208 to the test step 184, discussed above, where the backlog variable for VC,, L,, and die eligibility variable, e_t, are examined. If die backlog of VC, is at least one and if VC, is eligible, tiien control passes from the step 184 to the step 190, discussed above, where VC, is placed in die departure queue. Otiierwise, if die backlog number of cells for VC, is not at least one or if VC, is not eligible, tiien control passes from the step 184 to the step 186, discussed above, where the value of scd, is set to zero. Following die step 186, VC, enters die unscheduled state 144. Note diat, prior to transitioning to the step 184, it is possible to first run die cell admission check to determine if a cell has arrived since die last time the cell admission procedure has been run.

Referring to FIG. 6, a flow diagram 220 illustrates in detail die step 142 of FIG. 5 for setting up VQ. At a first step 222, r, and r₂ are set. The variables r, and r₂ represent data transmission rates. The variables r, can represent die average data transmission rate for die virtual data connection. The variables r₂ can represent die burst transmission rate for the data connection. The burst transmission rate is the maximum bandwiddi allowed for sending die data over die

virtual connection, VC,, by the requesting node. The system uses both the average rate and d e burst rate in a manner described in more detail hereinafter.

Following die step 222 is a step 223 where die weights, w,l-w,5 are set.

The weights are used in die calculation of the priority, P„ in a manner which is described in more detail hereinafter.

Following die step 223 is a step 224 where the quality of service, qos,, is set. The quality of service is provided by the sending node to die bandwiddi management unit and depends upon the nature of the data being transmitted on the virtual connections, VC,. The different possible values for qos, are discussed above.

Following die step 224 is a step 225 where c,l and c,2 are set to zero and c, _ml and c, ^ are set. The variables c,l and c,2 represent bandwidth credits diat are allocated to VC,. The credit variable c,l corresponds to the rate variable r,l and die credit variable c,l corresponds to die rate variable r,2. The variable c, _ml is die maximum value diat die variable c.l is allowed to reach. The variable c, ^ is the maximum value that the variable c,2 is allowed to reach.

The credit variables represent the number of cell slots allocated to VC, during die operation of the bandwiddi management unit. The credit variables are incremented whenever an amount of time has passed corresponding to die associated rate variable. For example, if r,l equals a rate corresponding to one hundred cells per second, than the credit variable c,l would be incremented once every one hundredtii of a second. The credit mechanism controls the amount of bandwiddi used by each of die VC's. Also, note that since r,l and r,2 can be different, tiien the credit variables c,l and c,2 can be incremented at different rates. Updating die credit variables is described in more detail hereinafter.

Following die step 225 is a step 226 where s, _m, and s, ^ are set. The variables s, _ml and s_{ι m2} represent die maximum value diat s, is allowed to reach. Following die step 226 is a step 227 where _,, die backlog of cells for VQ, is set to zero, and die variables 1, _m, L^,, and l_{m h}, which relate to 1, in a manner discussed in more detail hereinafter, are all set to predetermined constant values that vary according to die traffic parameters provided when VQ is initialized. Also at the step 227, the variables adπi_j, n,, and n,,,, which are discussed in more detail hereinafter, are also set

Following d e step 227 is a step 228 where die variable e„ which determines whether VC, is enabled, is set to one, thus enabling VQ, at least initially. Following die step 228 is a step 229 where die variable scd, is set to zero, thus indicating d at VC, is initially in the unscheduled state.

Following die step 229 is a step 230 where die variable t^ is set to zero.

The variable t,, indicates die value of die system time clock when the credit variables, c,l and c,2, were most recently updated.

Following die step 230 is a step 231 where the priority valuable, P„ is set to zero. Following die step 231 is a step 232 where die burst bit indicator, b, is initially set to zero.

Referring die FIG. 7, a diagram 240 illustrates in detail die cell admission step 146 of FIG. 5. The cell admission procedure shown in die diagram 240 is deemed "block mode" admission. Block mode admission involves potentially dropping a block of contiguous cells for a single VC in order to concentrate cell loss in that VC. That is, whenever a cell arrives for a VC having a full queue, the block mode admission algoritiim drops diat cell and may also drop a predetermined number of cells that follow the first dropped cell. The number of dropped cells is one of die initial traffic parameters provided when die VC is first established.

Processing for acceptance of cells begins at a current state 241 and transmission to a test step 242 when a new cell for VC, is provided to die bandwiddi management unit by the data node. If n, equals zero at die test step

242, control transfers to another test step 243. If it is determined at die test step

243 diat die number of backlog cells for VC,, 1,, is not less tiian the maximum allowable backlog of cells, 1, _m, then control transfers from the step 243 to a step

244 to determine if CLP equals zero. CLP is a variable indicating the cell loss priority of a cell. If a cell has a CLP of zero, that indicates diat die cell is relatively less loss tolerant (i.e., the cell should not be discarded if possible). A cell having a CLP of odier than zero is relatively more loss tolerant (i.e., could be discarded if necessary).

If CLP is not zero at die step 244, tiien control transfers to a step 245 where the newly arrived cell is discarded. Following die step 245, conu-ol transfer back to the current state 241. If it is determined at die test step 243 diat die number of backlogged cells, 1,, is less tiian the maximum allowable number of backlogged cell 1, _m, tiien control transfers from the step 243 to a step 246 to test if the value of CLP is zero.

If CLP equals zero at the step 246, then control transfers from the step 246 to a step 247 where the new cell is added to die cell buffer at the end of a queue of cells for VC,. Following the step 247 is a step 248 where die variable representing the backlog cells for VC,, l_j is incremented. Following die step 248, control transfers back to the current state 241 of VC,.

If at the step 246 it is determined diat CLP for die cell does not equal zero, then control transfers from the step 246 to a step 250 where adm_t is tested. The variable adπi_j is used to indicate which l_m variable, 1^ , or ^ __h, will be used. If at die test step 250 it is determined diat adm_t equals one, tiien control transfers from d e step 250 to a test step 251 to determine if me number of backlog cells for VC,, l_j, is less than the first backlog cell limit for VQ, Lj, _h. If so, tiien control transfers from the step 251 to a step 252 where the new cell is added to die queue for VC,.

Following die step 252 is a step 253 where die variable indicating the number of backlogged cells, L. is incremented. Following the step 253, control transfers back to the current state 241.

If it is determined at the step 250 diat adm, does not equal one, then control

transfers from d e step 250 to a step 255 where die number of backlogged cells for VC, is compared widi the second limit for die number of backlog cells for VC„ l_m ,. If at die test step 255 it is determined diat 1, is not less tiian L_^, ,, then control transfers from the step 255 to a step 256 where the newly arrived cell is discarded. Following die step 256, control transfers back to die current state 241.

If at the test step 255 it is determined diat die number of backlog cells for VQ, 1„ is less than the limit variable \_mJ, then control transfers from the step 255 to a step 258 where adm_; is set to one. Following die step 258 is die step 252 where the new cell is added to die queue of cells for VC,. Following die step 252, die variable representing the number of backlogged cells for VC,, _,, is incremented at die step 253. Following the step 253, control transfers back to current state 241.

If at die test step 251 it is determined diat the number of backlog cells for

VC, is not less than the limit variable l_{^ h}, tiien control transfers from the step 251 to a step 260 where adm, is set to zero. Following die step 260 is die step 245 where die newly arrived cell is discarded. Following die step 245, control transfers back to the current state 241. If at die test step 242 it is determined that n, does not equal zero, or if CLP does equal zero at die step 244, tiien control transfers to a step 262, where die newly arrived cell is discarded. Following die step 262 is a step 263 where n, is incremented. Following die step 263 is a test step 264, where n, is compared to die limit for n,, n_m. If at die step 264 n, is less tiian or equal to r , control transfers from die step 264 back to die current state 241. Otherwise, control

transfers to step 265, where n₄ is set to zero. Following the step 265, control transfers back to the current state 241.

Referring to FIG. 8, a flow diagram 280 illustrates die step 160 of FIG. 5 where die value of s, is initialized. Flow begins when VC, is in die unscheduled state 282. At a first test step 284, die values of c,l and c,2 are examined. The variable c,l and c,2 represent die credits for VC,. If die values of both of die credit variables, c,l and c,2, are greater tiian or equal to one, then control passes from the test step 284 to a step 286 where the value of s, is set to zero. Following the step 286, VC, enters a next state 288. Note that, if s, is set to zero at me step 286, then die next state 288 is die departure state 192, as shown on FIG. 5. This occurs because at die test step 164 of FIG. 5, die value of s, is checked to see if s, equals zero. If so, tiien VC, goes from die unscheduled state 144 directly to die departure state 192, as shown in FIG. 5 and as described above.

If at die test step 284 it is determined diat eitiier c,l or c,2 is less than one, then control transfers from the step 284 to a step 290 where s, is set to s, _ml. The variable s, _mι is a variable that represents die amount of time between cell transmissions for cells transmitted at a rate r,l.

Following die step 290 is a step 292 where ΔC is computed. The variable ΔC represents die amount of additional credit diat will be added to c,2 after an amount of time corresponding to s, has passed. The value of ΔC is computed by multiplying s, by r,2.

Following the step 292 is a test step 294 which determines if the existing value of die credit variable, c,2, plus die value of ΔC, is less tiian one. Note tiiat the variable c,2 is not updated at this step. If the sum of c,2 and ΔC is not less tiian one, then control transfers from the step 294 to die next state 288. Note that, in this case, die next state will be die eligibility state 168 shown in FIG. 5 because s, will not equal zero at die step 164 of FIG. 5.

If die sum of c,2 and ΔC is less tiian one at die step 294, then control transfers from this step 294 to a step 296 where the value of Sj is set to s_{ι m2}. The variable s _m2 represents the amount of time between cell transmissions for cells transmitted at a rate r,2. Following the step 296, control transfers to the next state 288, discussed above.

Referring to FIG. 9, a flow diagram 300 illustrates the reset s_{ value step 204 shown in FIG. 5. At a first test step 302, the burst bit variable for VQ, b„ is tested. The burst bit variable, b,, indicates that VC, should transmit in burst mode. Burst mode is used for a virtual connection where die cells of a virtual connection should be transmitted as close togetiier as possible. That is, the data node sets the virtual connection for burst mode for data traffic that should be transmitted as close togetiier as possible. For example, although an e-mail message may have a low priority, it may be advantageous to transmit the entire e-mail message once the first cell of the e-mail message has been sent.

If the burst bit for VQ, b„ is determined to be set at die step 302, control passes from the test step 302 to a test step 304 where the credit variables, c,l and c,2, and die backlog variable, 1,, for VC, are tested. If die credit variables and die backlog are all greater tiian or equal to one at the step 304, tiien control passes from tiie step 304 to a step 306 where s, is set to zero. Following step 306, VC, enters the next state 308. Note that, since s, is set to zero at the step 306, the next state for VC, will be die departure state 192 shown in FIG. 5.

If die burst bit indicator, b„ is set at die step 302 but VQ does not have enough bandwiddi credits or die backlog 1, is zero at die step 304, tiien control passes from the step 304 to a step 310 where s, is set equal to s, _ml. Note that the step 310 is also reached if it is determined at die test step 302 diat die burst bit indicator, b_j does not equal one (i.e., VC, is not transmitting in burst mode).

Following the step 310 is a step 312 where ΔC is computed in a manner similar to at illustrated in connection with the step 292 of FIG. 8. Following the step 312 is a step 314 where it is determined if sum of die credit variable c,2 and

ΔC is less tiian one. If not, then control passes from the step 314 to die next state 308. The test step 314 determines if die credit variable c,2 will be greater tiian one after an amount of time corresponding to s_t has passed (i.e., after the s_s counter has timed out). If not, tiien c_t2 plus ΔC will be less than one when s, times out and control transfers from the step 314 to a step 316 where s, is set equal to s_t ^ Following the step 316, VC, enters the next state 308.

Referring to FIG. 10, a flow diagram 320 illustrates the steps for updating c,l and c,2, d e credit variables for VC,. Updating c,l and c,2 occurs at die step 158 and die step 202 shown in FIG. 5, discussed above. Note diat die credit variables are set according to current system time but that, for the step 202 of FIG. 5, the time used is the beginning of the next cell time slot in order to account for die amount of time it takes to transmit a cell at the steps preceding the step 202.

Processing begins at a current state 322 which is followed by a step 324 where a value for Cjl is computed. At die step 324, c,l is increased by an amount equal to the product of r,l (one of die rate variables for VQ) and die differences between die current system time, t, and die time at which the credit variables, l and C_j2, were last updated, t, _n. Following the step 324 is a test step 326 to determine if c_sl is greater tiian c_ilm. The variable c,_lm is the maximum value that the credit variable c,l is allowed to equal and is set up at die time VC, is initialized. If at the test step 326 c,l is greater man c,_lm, then control transfers from the step 326 to a step 328 where c,l is set equal to c,_lm. The steps 326, 328 effectively set c,l to the maximum of either c,l or c_ilm. Following the step 328 or die step 326 if c,l is not greater tiian c,_lm is a step 330 where a value for c,2 is computed in a manner similar to die computation of c,l at the step 324. Following the step 330 are steps 332, 334 which set c,2 to the maximum of either c,2 or c_l2rn in a manner similar to the steps 326, 328 where c,l is set equal to the maximum of c,l and c_tlm.

Following die step 334, or the step 332 if c,2 is not greater tiian c^, is a step 336 where t,,, is set equal to t, the current system time. The variable t_m represents the value of the system time, t, when the credit variables, c,l and c,2, were last updated. Following die steps 336 is a step 338 where the VC, enters the next appropriate state.

Referring to FIG. 11, a flow diagram 340 illustrates the step 174 of FIG. 5 where s, is updated by ΔS. The step 174 of FIG. 5 is executed whenever die data node associated widi die bandwiddi management unit requests changing die delay of sending data via VC, by changing the value of s, for VC,.

Processing begins with VC, in an eligibility state 342. When die data source node provides VC, with a command to update s,, control transfers from the step 342 to a test step 344 to determine if the value of ΔS, provided by die associated data node, equals zero. If so, control transfers from the step 344 back to the eligibility state 342 and no processing occurs. Otiierwise, if ΔS does not equal zero, then control transfers from the test step 344 to a step 346 where VC, is located in the eligibility queue. Note diat, as shown in FIG. 5, the step 174 where S_j is updated only occurs when VC, is in die eligibility state.

Following die step 346 is a test step 348 where it is determined if ΔS is greater tiian zero or less than zero. If ΔS is greater tiian 0, then control transfers from die step 348 to a step 350 where ΔS is added to s_;. Following the step 350 is a step 352 where VC, is repositioned in die eligibility queue according to die new value of s_t. Note that, as discussed above in connection with FIG. 5, the position of VQ in the eligibility queue is a function of the values of s_s and qoS_j.

If at the test step 348 it is determined diat ΔS is less tiian zero, then control transfers from the step 348 to a step 354 where the credit variables, c,l and c,2, are examined. If at die test step 354 it is determined diat either c_tl or Cj2 is less than one, then control transfers from the step 354 back to the eligibility state 342 and no update of s_t is performed. The steps 348, 354 indicate diat die value of s, is not decreased if eitiier of the credit variables is less tiian one.

If at die step 354 it is determined diat die credit variables, c_tl and c,2, are botii greater than or equal to one, then control transfers from the step 354 to a step

356 where the value of s, is incremented by die amount ΔS. Note diat, in order to reach die step 356, ΔS must be less tiian zero so that at die step 356, die value of s_s is in fact decreased.

Following die step 356 is a step 358 where die value of ^ is compared to s(t). If die value of S_j has been decreased at die step 356 by an amount that would make Sj less than s(t) then control transfers from the step 358 to a step 360 where s, is set equal to s(t). The steps 358, 360 serve to set the value of s, to the greater of either s, or s(t).

Following the step 360 or following the step 358 if s_{ is greater than or equal to s(t), control transfers to a step 362 where VC, is repositioned in die eligibility queue. As discussed above, die position of VC, in the eligibility queue is a function of the values of s, and qos,.. Following e step 362, VC, returns to die eligibility stage 342.

Referring to FIG. 12, a flow diagram 370 illustrates the compute steps 162, 206 of FIG. 5 where the values of c,l, c,2, and P, are computed. Processing begins at a current state 372 which is followed by a step 374 where the value of c,l is computed. The value of c,l equals die value diat die credit variables will have when s, times out. That is, c,l equals die future value of die credit variables c,l when VC, is removed from die eligibility queue. Since, as discussed in more detail hereinafter, P, is a function of die credit variables, c,l and c,2, then calculating die future values of c,l and c,2 is useful for anticipating die value of P, when VC, is removed from die eligibility queue and placed in die departure queue.

The value of die priority variable, P,, is determined using die following equation:

P, = w,l*c,l + w,2*c,2 + w,3*r,l + w,4*r,2 + w,5*L. P_; is a function of the credit variables, die rate variables, and the number of cells backlogged in die queue. The node requesting initialization of VC, specifies the weights, w,l-w,5, for the VC. Accordingly, it is possible for die requesting node to weight die terms differently, depending upon the application. For example, if the data is relatively bursty, then perhaps w,2 and w,4 (die weights for die terms corresponding to the burst data rate) are made larger tiian w,l and w,3 (die weights for die terms corresponding to die average data rate). Similarly, if it is desirable to minimize die backlog for die data, tiien w,5 (the weight corresponding to die number of backlogged cells) may be made relatively large. It may be desirable for some applications to assign zero to some of the weights.

At the step 374, the value of c,l is calculated by adding die current value of the credit variables c,l and die product of r,l and s,. Note that, if the credit variables were updated for each iteration while VC, was in the eligibility queue, then c,l would equal c,l when s, equalled zero. Note diat c,l is not updated at the step 374.

Following the step 374 is a test step 376 where c_sl is compared to c,, _m. If c,l is greater tiian c,, _m (the maximum value at c,l is allowed to take on) tiien control transfers from the test step 376 to a step 378 where the value of c,l is set equal to the value of c„ _m. The steps 376, 378 serve to set the value of c,l to the maximum of c,l and c,, _m. Note diat for die steps 376, 378, c,l is not updated.

Following die step 378 or following die step 376 if c,l is not greater tiian c,_{ι m} are steps 380-382, which set the value of c,2 in a manner similar to setting die value of c,l, described above.

Following die step 382 or following the step 381 if c,2 is not greater than ^ _m is a step 384 where the weights w,l, w,2, w,3, w,4, and w,5 are fetched. As discussed above, die weights are used to calculate the priority, P,.

Following the step 384 is a step 386 where the priority for VQ, P„ is calculated in die manner shown on die flow diagram 370. Note mat the weights are specified at die time that VC, is initialized and diat die importance of each of the terms used in die calculations of P,, can be emphasized or deemphasized by setting the value of the weights. The values of one or more of the weights can be set to zero, thus causing a term not to be used in die calculation of die priority. Note that P, is updated, P=P,+w,5, after each VC, cell admission while VC, is in die eligibility state, as discussed above. This allows for a piecewise computation of P, that decreases the amount of computation required to compute all of die priorities for all of the VC's within one time slot when the VC's are transferred to the departure queue (die "B" queue).

Referring to FIG. 13, a flow diagram 390 illustrates in detail die step 180 of FIG. 5 where the priority variable, P,, is updated. Note diat P, is updated only in response to die value of s, being updated at die step 174. Otherwise, the value of P, remains constant while VC, is in the eligibility state. That is, since P, is a function of c,l, c,2, r,l, r,2 and L., and since c,l and q2 are a function of s„ then P, only changes when and if s, changes at die step 174 shown in FIG. 5.

Flow begins at a current state 392 which is followed by a step 394 where s, is accessed. Following the step 394 are steps 396-398 where c,l is computed in a manner similar to mat illustrated in FIG. 12. Note, however, diat at die step 396, ^ is subtracted from s, since, at is stage of the processing, s(t) has already been added to s, at die step 166 shown in FIG. 5, and since c,l is being predicted when s, times out.

Following the step 398 are steps 400-402 where c,2 is computed in a manner similar to the computation of c.l. Following the step 402 or following the step 401 if c,2 is not greater than c_{α m} are the steps 404, 405 where die weights are fetched and P, is computed in a manner similar to that illustrated in connection with FIG. 12, described above.

Referring to FIG. 14, a schematic diagram 420 illustrates in detail one possible hardware implementation for die bandwiddi management unit. An input 422 connects die bandwidth management unit 420 to the source of the data cells.

The input 422 is connected to a cell pool 424 which is implemented using memory. The cell pool 424 stores all of the cells waiting for departure from die bandwidth management unit. In the embodiment illustrated herein, die size of die cell pool 424 equals the number of VC's that die system can accommodate multiplied by the maximum allowable backlog per VC. For example, for a bandwiddi management unit designed to handle 4,096 VC's widi a maximum backlog lengtii of 30 cells, die size of die cell pool is 4096 x 30 cells.

A cell pool manager 426 handles die cells witiiin the bandwidth management unit 420 in a manner described above in connection with die state diagrams of FIG's 5-13.

A parameter buffer 428 is used to handle all of die variables associated widi each of die VC's and is implemented using memory. The size of die parameter buffer 428 is die number of variables per VC multiplied by the maximum number of VC's handled by die bandwiddi management unit 420.

A parameter update unit 430 is connected to die parameter buffer 428 and updates die variables for each of the VC's in a manner described above in connection with the state diagrams of FIG.'s 5-13.

A cell clock 432 computes the value of s(t) which, as described above, is used for determining when a VC transitions from the eligibility queue to the departure queue. The cell clock 432 also provides die system time, t.

An eligibility queue 431 is implemented using memory and stores die VC's that are in the eligibility state. As discussed above in connection widi FIG.'s 5-13, VC's in die eligibility queue are sorted in order of the values of s and qos for each

VC. A comparator 434 compares die value of s(t) from the eligibility cell clock 432 with the value of s for each of the VC's and determines when a VC should transition from the eligibility state to the departure state. Note diat, as discussed above in connection with FIG.'s 5-13, VQ transitions from the eligibility state to the departure state when s, > s(t).

A departure queue 436 is implemented using memory and stores die VC's diat are awaiting departure from die bandwiddi management unit 420. As discussed above in connection with FIG.'s 5-13, die departure queue 436 contains VC's that are sorted in order of quality of service (qos) for each VC and die priority that is computed for each VC. A select sequencer unit 433 selects, for each VC, which of the two queues 431, 436 to place each VC according to die algoritiim described above and shown in FIG.'s 5-13.

A priority computation unit 438 computes the priority for each of die VC's during each iteration. Note diat, as discussed above in connection with FIG.'s 5-13, once a VC has entered die eligibility queue, the priority can be updated by simply summing die value of the priority variable and the product of the weight w₅ after each cell arrival is admitted.

The device 420 shown in FIG. 14 can be implemented in a straightforward manner using conventional VLSI architecture. The cell pool 424, eligibility queue 431, departure queue 436, and parameter buffer 428 can be implemented as memories. The cell pool manager 426, die parameter update unit 430, die select sequencer 433, the priority computation unit 438 and the comparator 434 can be implemented using VLSI logic to implement die functionality described above in connection with FIG.'s 5-13. That is, one of ordinary skill in die art can implement the bandwiddi management unit 420 by using die state diagrams and current custom VLSI design techniques. Note also that, based on die description contained herein, one of ordinary skill in die art could implement die bandwiddi management unit 420 in a variety of other manners, including entirely as software on a very fast processor or as a conventional combination of hardware and software.

Although the invention has been shown and described witii respect to exemplary embodiments tiiereof, it should be understood by those skilled in die art that various changes, omissions and additions may be made therein and thereto, without departing from me spirit and the scope of the invention.

Claims

WHAT IS CLAIMED IS:

1. A bandwiddi management system, for managing a plurality of virtual data connections within a communications network, comprising:

an input for receiving data cells, wherein each cell is associated widi a particular one of the virtual connections; a cell pool, coupled to said input, for storing said cells; a first queue containing particular ones of die virtual connections, wherein a relative position of a virtual connection in said first queue is determined by an eligibility value that varies according to an anticipated data rate associated with the particular virtual connection and according to an amount of time that the particular virtual connection has been in said first queue; a second queue, coupled to said first queue and containing particular other ones of the virtual connections, wherein a relative position of a virtual connection in said second queue varies according to a predetermined quality of service that is assigned to each of die virtual connections; and an output, coupled to said second queue and said cell pool, for transmitting a cell from said cell pool corresponding to a virtual connection at the front of said

second queue.

2. A bandwidth management system, according to claim 1, wherein virtual connections having equal eligibility values are ordered in said first queue according to said predetermined quality of service.

3. A bandwidth management system, according to claim 1, wherein virtual connections having equal quality of service values are ordered in said second queue according to a priority assigned to each of said virtual connections.

4. A bandwiddi management system, according to claim 3, wherein said priority varies according to one or more anticipated data rates of each of the virtual connections.

5. A bandwiddi management system, according to claim 3, wherein said priority varies according to one or more credit variables assigned to each of die virtual connections, said credit variables being indicative of allocated time slots provided to each of die virtual connections.

6. A bandwiddi management system, according to claim 3, wherein said priority varies according to a backlog of cells for each virtual connection that are awaiting transmission.

7. A bandwiddi management system, according to claim 3, wherein said priority varies according to one or more anticipated data rates of each of die virtual connections, one or more credit variables assigned to each of the virtual connections, said credit variables being indicative of allocated time slots provided to each of die virtual connections, and to a backlog of cells for each virtual connection that are awaiting transmission.

8. A bandwiddi management system, according to claim 7, wherein said data rates, said credit variables, and said backlog are weighted prior to determining said priority.

9. A bandwiddi management system, according to claim 1, wherein said system is one of: a pacing unit and an enforcement unit, wherein said pacing unit receives cells from a data source node and provides cells to a communication link, and wherein an enforcement unit receives data from a communication link and provides data to a data sink node.

10. A bandwiddi management system, according to claim 1, further comprising: means for establishing a virtual connection by specifying initial traffic

parameters; and means for dropping cells diat are received at a rate diat exceeds diat specified by said initial traffic parameters.

11. A bandwiddi management system, according to claim 1, wherein said predetermined quality of service has four possible values.

12. A bandwidth management system, according to claim 11, wherein virtual connections having equal quality of service values are ordered in said second queue according to a priority assigned to each of said virtual connections.

13. A bandwiddi management system, according to claim 12, wherein said priority varies according to one or more anticipated data rates of each of die virtual connections, one or more credit variables assigned to each of die virtual connections, said credit variables being indicative of allocated time slots provided to each of die virtual connections, and to a backlog of cells for each virtual connection that are awaiting transmission.

14. A bandwiddi management system, according to claim 13, wherein said data rates, said credit variables, and said backlog are weighted prior to determining said

priority.

15. A bandwiddi management system, according to claim 1, further comprising:

a burst bit indicator for indicating whether a virtual connection associated with a cell that has been received by said input should be placed directly in said second queue.

16. A bandwidth management system, according to claim 15, wherein virtual connections having equal quality of service values are ordered in said second queue according to a priority assigned to each of said virtual connections.

17. A bandwiddi management system, according to claim 16, wherein said priority varies according to one or more anticipated data rates of each of die virtual connections, one or more credit variables assigned to each of die virtual connections, said credit variables being indicative of allocated time slots provided to each of die virtual connections, and to a backlog of cells for each virtual connection that are awaiting transmission.

18. A bandwiddi management system, according to claim 17, wherein said data rates, said credit variables, and said backlog are weighted prior to determining said priority.

19. A metiiod of managing a plurality of virtual data connections within a communications network, comprising the steps of: receiving data cells, wherein each cell is associated with a particular one of die virtual connections; storing die received cells in a cell pool; providing a first queue containing particular ones of die virtual connections, wherein a relative position of a virtual connection in the first queue is determined by an eligibility value that varies according to an anticipated data rate associated witii the particular virtual connection and according to an amount of time diat die particular virtual connection has been in the first queue; providing a second queue, coupled to die first queue and containing particular odier ones of die virtual connections, wherein a relative position of a virtual connection in the second queue varies according to a predetermined quality of service diat is assigned to each of the virtual connections; and transmitting a cell from the cell pool corresponding to a virtual connection at the front of die second queue.

20. A metiiod, according to claim 19, further comprising the step of: ordering virtual connections having equal eligibility values in the first queue according to die predetermined quality of service.

21. A method, according to claim 19, further comprising the step of: ordering virtual connections having equal quality of service values in the second queue according to a priority assigned to each of die virtual connections.

22. A metiiod, according to claim 21, further including die step of: varying die priority according to one or more anticipated data rates of each of die virtual connections.

23. A method, according to claim 21, further including die step of: varying die priority according to one or more credit variables assigned to each of die virtual connections, wherein die credit variables are indicative of allocated time slots provided to each of die virtual connections.

24. A metiiod, according to claim 21, further including the step of: varying the priority according to a backlog of cells, for each virtual connection, diat are awaiting transmission.

25. A method, according to claim 21, further including die step of: varying die priority according to one or more anticipated data rates of each of die virtual connections, one or more credit variables assigned to each of die virtual connections, wherein the credit variables are indicative of allocated time slots provided to each of die virtual connections, and according to a backlog of cells, for each virtual connection, that are awaiting transmission.

26. A method, according to claim 25, further including die step of: varying die priority according to weights assigned to die data rates, die credit variables, and die backlog.

27. A metiiod, according to claim 19, further comprising the steps of establishing a virtual connection by specifying initial traffic parameters; and dropping cells that are received at a rate diat exceeds diat specified by die initial traffic parameters.

28. A method, according to claim 19, further including die step of: assigning one of four possible values die predetermined quality of service.

29. A metiiod, according to claim 28, further comprising the step of: ordering virtual connections having equal quality of service values in the second queue according to a priority assigned to each of the virtual connections.

30. A method, according to claim 29, further including die step of: varying die priority according to one or more anticipated data rates of each of die virtual connections, one or more credit variables assigned to each of die virtual connections, wherein the credit variables are indicative of allocated time slots provided to each of die virtual connections, and according to a backlog of cells, for each virtual connection, that are awaiting transmission.

31. A method, according to claim 30, further including die step of: varying die priority according to weights assigned to die data rates, die credit variables, and die backlog.

32. A metiiod, according to claim 19, further comprising the step of:

placing a virtual connection associated witii an admitted cell directly on the second queue in response to a burst indicator for the virtual connection being set.

33. A method, according to claim 32, further comprising the step of: ordering virtual connections having equal quality of service values in die second queue according to a priority assigned to each of die virtual connections.

34. A metiiod, according to claim 33, further including die step of: varying the priority according to one or more anticipated data rates of each of die virtual connections, one or more credit variables assigned to each of die virtual connections, wherein the credit variables are indicative of allocated time slots provided to each of die virtual connections, and according to a backlog of cells, for each virtual connection, that are awaiting transmission.

35. A method, according to claim 34, further including die step of: varying die priority according to weights assigned to die data rates, die credit variables, and the backlog.