CA2334301C - Method for dynamic rate allocation in a communication network, in particular a high rate network - Google Patents

Abstract

The invention concerns a method for dynamic rate allocation in a communication network, in particular a high rate network such as an ATM (Asynchronous Transfer Mode) network. Said method is designed to be implemented at the network nodes, and consists in allocating, to each of the sources whereof the connections converge on one network node and in response to a rate request from said source, a cell rate so as to share the maximum band provided by said node output link.

Description

Method for Dynamic Rate Allocation in a Communication Network, in Particular a High Rate Network.

FIELD OF THE INVENTION
The invention concerns a method for dynamic rate allocation in a communication network, in particular, a high rate network such as an ATM (Asynchronous Transfer Mode) network.
BACKGROUND OF THE INVENTION
ATM networks were designed to support the flow of multimedia information, such as data, sound, images and video. ATM technology permits the same infrastructure to transmit data flow coming at different rates and with different performance requirements. In order to offer these characteristics, ATM networks are anticipated to enable resource management and guarantee service quality.

Before being established, a network connection request, specifying the desired characteristics, in particular, the rate it will reserve and the necessary quality of service, must be made. As needed, the connection states, in its request, the minimum rate the network must guarantee it. If the network decides that it has sufficient resources to meet the request, it establishes the connection and also guarantees the rate and connection quality for the requested service.

Different classes of service have been defined which in turn, according to the type of data source and performance and quality required from the corresponding connection, define the connection's traffic control protocol. The classes usually considered are: the class known as CBR (Constant Bit Rate), the class known as VBR (Variable Bit Rate), the class known as ABR
(Available Bit Rate), and a class related to ABR, the class known as ABT (ATM
Block Transfer).

To simplify, it can be said that in the CBR class, high-quality service is specified, whereas in the ABR class, there are no explicit requirements for service quality with respect to transfer time or gigs. In the VBR class, a certain number of control parameters are defined at the time of the connection, and once defined, these limits are used by the network to guarantee service quality.
The mechanisms which are generally implemented for the service classes ABR and ABT, the only ones considered here, are the following: for each active connection, a specialized bit, hereinafter referred to as the resource management bit, will run the entire route followed by the bits using the connection. When a network node receives this bit, a protocol is implemented to verify that the rate allocated between the node and the connection does not exceed a value which is determined by the other connections at the time. If this is the case, the node will allocate a new rate for the connection of which the value would be recorded in the resource management bit.
When the resource management bit arrives at the end of the route followed by the user bits, it is returned to the source of the connection in question with the lowest rate authorized along the length of the route. The source must thus adjust its rate to this new value.

An allocation method is planned for implementation along the nodes of a network. It permits the allocation of a bit rate to each source of which the connections converge at a network node, after a rate request has been made by the source. The allocated bit rate will allow the maximum band offered by the output link of the said node to be shared.

A sample description of the traffic control mechanisms implemented for the ABR
class can be taken from F. BONOMI and K.W. FENDICK in an article entitled "The rate-based flow control framework for the Available bit Rate Atm service" which was published in the March/April 1995 issue of the IEEE Network.

Known rate allocation methods are generally implemented for connections which support data and use FIFO-type queues in the nodes, which cannot be envisaged for video transmission as, since video rate specifications are necessarily high, transmission delays cannot be too great, etc.

2 The goal of this invention is thus to propose a rate allocation method which enables equitable sharing of the band used by different real-time sources, notably video sources.
SUMMARY OF THE INVENTION
To that end, a method according to this invention is proposed which will, over successive cycles, do the following:

a) at the beginning of each cycle, assign a rate to each connection; and b) during the cycle, on the one hand, allocate the assigned rate to each connection whose required rate is greater than that assigned, that connection is thus accounted for and marked as clipped; and. on the other hand, allocate to each connection whose required rate is less than that assigned, either the rate already allocated in the preceding cycle if in that previous cycle the said connection had not been marked as a clipped connection, or the rate required if in the preceding cycle the connection had been marked as a clipped connection. The difference between the assigned rate and the allocated rate is thus accounted for as unallc-cated rate; and c) at the end of the cycle, calculate a new rate to assign to each of the connections in question for the following cycle, based on the unallocated rate accounted for and the number of clipped connections.

Another characteristic of the invention would be to assign to each source, during an initial cycle, a rate which corresponds to an. equitable share for each of the maximum band sources offered by the node output link.

Another characteristic of the invention would be to calculate the new rate in order to assign to each of the clipped connections in the following cycle, sharing the unallocated rate equally among the clipped connections.

3 Another characteristic of the invention would be to calculate the new rate to assign to each of the connections in the following cycle, sharing equally the unallocated rate, minus the excess assigned rate, among the said clipped connections.

Another characteristic of the invention would be to ensure that the duration of each cycle is greater than the renegotiation period implemented by the network equipment.

Another characteristic of the invention would be to ensure that the duration of each cycle is greater than the network equipment renegotiation period increased by the maximum amount of time taken by the resource management bits of the said connections to make the complete trip from the source in question to the other end of the network and back again to the source.

Another characteristic of the invention would be to count the requests from each source, examining only requests that have an order number.

Another characteristic of the invention would be to mark a connection after it receives the request from the corresponding source, erasing the mark at the beginning of each cycle, and examining only requests from sources whose connections have already been marked.

Another characteristic of the invention would be to weigh the rates assigned to each source. Each weight factor would depend on the minimum guaranteed rate requested by the corresponding source. For example, the weight factor could be equal to the minimum guaranteed rate requested by the source divided by the total minimum guaranteed rate requested by all sources. The rate assigned to a source may thus be equal to a rate common to all the sources which are multiplied by the weight factor.

Another characteristic of the invention would be when a request is received from a source whose connection has not been marked as clipped in the previous cycle, and when the rate now

4 requested is greater than the rate which was allocated in the previous cycle, to interrupt the current cycle and begin a new cycle with this request, with the rate assigned to each source being either the current rate of the cannection in question or the rate shared equally among all sources, whichever is greater.

Another characteristic of the invention would be to assign a cycle identification number to each connection.

In accordance with a first aspect of the present invention there is provided a method for dynamic rate allocation for implementation in network nodes, each network node having a plurality of sources, each source having a connection to a network node, the method comprising:
allocating to each connection, and in response to one of a plurality of periodic rate requests made by a source corresponding to the connection, a cell rate selected from a lesser of a request rate and a residual rate after allocation of requested rates to all the other sources, wherein, during successive cycles:

a) at the beginning of each cycle of receipt of requests originating from the sources corresponding to the connections, considering and assigning an assigned rate (Rl, Ri) to each connection; and b) during the cycle, b1) for each connection having a required rate (Rq) which exceeds the assigned rate (R1, Ri) for the cycle, allocating the assigned rate (R1, Ri) to that and marking it as a clipped connection for the cycle;
b2) for each connection having a required rate (Rq) which is less than the assigned rate (Rl, Ri), determining whether that connection is marked as a clipped connection from a previous cycle, and if so, allocating the difference between the value of the assigned rate and the value of the rate allocated as a non-allocated rate value (0); and if not, allocating the rate already allocated to the connection in previous cycle; and

5 b3) after a rate has been allocated to each connection calculating the number (m) of clipped connections, and calculating a new value for the assigned rate (R1, Ri) for the subsequent cycle using the following relation:

R1 = R1 + C-N=R1 +p m wherein C is the rate value of the maximum band offered by the outgoing link at the node, N is the number of active connections on the node, 0 is the non-allocated rate and m is the number of clipped connections.

BRIEF DESCRIPTION OF THE DRAWINGS
The above-mentioned characteristics of the invention, and others, will become more clear by reading the following description of an example of this invention's implementation accompanied by the attached <irawing which illustrates the invention's method as illustrated by an arbitrary example.

DESCRIPTION OF THE PREFERRED EMBODIMENTS
According to the invention, the method is based in each network node. For each active communication which enters the node in question, the node holds a Reg register which stores the rate that is allocated to each communication. At intervals of time designated by T, the source of each communication sends an RM resource management bit onto the communication path. At the beginning, the RM bit can-ies the rate requested by the source. At each node, the RM
management bit rate and the rate stored in the node Reg register are compared.
If the rate stored in the Reg register is less than. the RM management bit rate, then the RM
management bit rate is changed to the rate stored in the Reg register. On the other hand, if the RM
management bit rate is equal to or greater than the stored rate, it is not changed.

Once it arrives at the end of the network, the RM management bit returns to the source. It has the lowest communication rate al:located by the network which is therefore authorized along the entire length of the communication path. Thus the source adjusts its user bit rate to the new rate.
5a Note that we call the bits which carry data 'user bits', for example, image and sound data sent by the source to a requesting party.

This invention's method is thus concerned with the allocation of communication rates. This allocation is dynamic in the sense that there is a rate renegotiation carried out over the RM
resource management bits' T intervals. Thus, for each communication which arrives at a network node, the node will receive, at T intervals, one and only one renegotiation request and respond to it, using this invention's procedure, by allocating a user bit rate to the user bits.

The invention's method is thus used over successive cycles of T intervals. We will now describe the various steps of the method, as they appear in a cycle.

If we call the maximum band which can support the output link of the node in question 'C', then distributing the rate equally over the N connection sources which meet at this node allocates to each source a rate of C/N. Therefore, over an initial cycle, this value ofRi =
C/Nis placed in each of the Reg registers linked to the active communications.

During each cycle, when the node receives the RM resource management bit sent by the source whose connection is converging on the node, two cases may arise.

In the first case, the Rq rate required by the source and carried by the RM
bit is less than the R1 rate. Thus an RD register accounting for the A quantity of band which has not been allocated during the current cycle is increased by the amount equal to R1 - Rq. In addition, the communication is marked as being unable to increase its rate beyond the Rq value.

In the other case, the Rq required rate is greater than the Rl rate. The source request is said to be clipped at the R 1 rate and is thus marked as a result. In addition, the count on the m counter (which counts the number of clipped requests) is increased by increments.

6 At the end of a cycle, when all the sources have transmitted their resource management bit RM, that is to say, after a T interval corresponding to the time it takes for the sources to send the RM
bits the situation is as follows.

The communications whose Rq rate requests are less than the R 1 rate are marked as being unable to increase their rate beyond that Rq rate. The others will be allocated a rate which is for the moment equal to the Rl rate. The total of the band left unallocated during the cycle is stored in the L register and the number of clipped communications is stored in the m counter.

In this way, we know the number of connections which need to increase their rate and the remaining band available for that increase. With this invention, rate increases can be authorized for clipped connections in the next cycle. Rates would be increased by a value corresponding to the unallocated band of the node output link divided by the number m of the clipped connections.

In the following cycle we can also re-implement the invention's method with a new value ofRl and authorize rate increases only for those connections previously marked as clipped.

We will now explain the invention's method first by using an arbitrary example as an illustration. In this example, there are three communications, C1, C2, and C3 (see figure 1). The first C1 requires a rate of two units, C2 requires a 6-unit rate and C3 requires nine units. The maximum capacity C of the node output link is 15 units (C = 15).

After the initial cycle CYI, each communication will be assigned an equal share of the R 1 rate, the value of which is thus R 1= C/N where N is the number of active communications converging on the node in question. Here N= 3. Thus Rl = 5. The total assigned rate is N-R1, or 15 units.

Using the terminology adopted here, the assignment of a rate to a communication is not the allocation. For a connection which was previously clipped, the assigned rate is the maximum rate

7 which the node can allocate it. On the other hand, the assigned rate for a connection which has not been previously clipped bears no relation to the allocated rate. The difference between the assigned rate and the allocated rate is accounted for in unallocated band.

This communication C1 is allocated two units and there are three unallocated units which are accounted for in the RA register (L = 3). Communications C2 and C3 were clipped at five units, and are thus allocated only 5 units. For communications C2 and C3 only, the allocated rate is equal to the assigned rate.

At the end of the cycle, there are therefore A = 3 and m = 2 (number of clipped communications = 2).

In the next cycle CY2, the three units which were unallocated in the previous CYl cycle will be distributed equally to the clipped communications C2 and C3, L/m = 3/2 = 1.5 units per conununication. To do this, we allocate an equally shared R1 rate of (5 + 1.5) units, or 6.5 units, to each active communication on the node in question. The total assigned rate is thus N- R1-=
6.5x3=19.5units.

The Cl communication is allocated two units and there remains, with respect to assigned units, 4.5 unallocated units. Communication C2 is allocated six units in compliance with its request, and there remain 0.5 unallocated units which, added to the 4.5 units already unallocated from conununication C l, totals five unallocated units.

Communication C3 is allocated 6.5 units. Communication C3 is the only communication to have been clipped during this cycle. At the end of the cycle therefore we have A 5 and m=1.

8 To calculate the band to be shared in the next cycle, we must subtract the assigned rate which exceeds the maximum capacity C of the node output link from the five unallocated units, that is (N - Rl - C) =(19.5 - 15) = 4.5 units. The band to be shared in the following cycle is thus equal to: 5 - 4.5 = 0.5 units allowing the calculation of the new R1 value which now equals 6.5 + 0.5 =
7 units.

In the CY3 cycle, each active communication will be assigned 7 units, which corresponds to the new R1 rate. The total assigned rate is thus 3 x 7 units, or 21 units. You will notice that at the end of the CY3 cycle, 0 will be equal to 6. The band to be shared in the next cycle will thus be A -(N=R1-C)=0.

The process continues in this way from cycle to cycle.
We will now formalize the invention's method.

The band available for all communications is the band available-0n the node output link, which is C.

The total equally assigned rate, to the rate of each cycle, is called Da and its value is equal to N- R 1, from which:

Da=N=Rl However, the total assigned Da may be greater than the node output link's maximum capacity C.
The result is an assigned rate which exceeds the maximum capacity of the node output link. The excess rate De thus has the value:

9 De=Da-C=N-RI-C.

Therefore the relationship can be expressed in this way:
C-Da-De =Da-(N=Rl - C) = Da + (C- N - RI) At the end of each cycle, the total assigned rate Da was, in the first portion, allocated, and in the second portion, unallocated. In the allocated portion there is, in one portion, the rate effectively allocated to the clipped connections, or m.Rl, where m represents the number of clipped connections and R1 represents the rate assigned to each communication, and in the other portion, the total rate allocated to unclipped connections, or LZ 1uncripped rate.] The second, unallocated portion, was accounted for in the Rni register. This can be expressed in the following manner:
Da = m= R 1+ E rate + A

unclipped And so, replacing the previous expression:
C-m-Rl+(C-N-R1)+E rate + A
unclipped Which can also be written as follows:
C = m =(R1+ C-N-Rl+p) + Yrate m unclipped Through this relationship, it can be seen that if, in the next cycle, the rate included in parentheses is attributed to the connections which were previously clipped, the band will be shared equally, and fully used.

This invention will proceed in the following manner. In the next cycle, the invention's method will be implemented again, authorizing only those connections which were clipped to increase their rate again. To do this, the R1 value is modified and now equals:

R 1= R 1+(C-N=R 1) +p m Thus, at the beginning of each cycle, there are a certain number of connections which are not authorized to increase their rate whereas others are authorized to do so up to a value corresponding to the value stored in the R 1 register. Since the connections are examined individually in the process explained above, at the end of the cycle, the available unallocated rate A and the number m of clipped connections may be determined. From that point, the new value ofRi which to be used for the examination of the next cycle is recalculated.
The process continues to function in this way continually from cycle to cycle.

With this invention, if a connection that was not clipped in the previous cycle wishes to increase its rate in the current cycle, the current cycle is interrupted. The rate stored in the R1 register is modified and a new cycle is started based on this connection. The new Ri rate equals the rate previously allocated to the connection in question or the equally shared rate C/N, whichever is greater, C being the rate available on the node output link in question, and N
being the number of active connections at the time the new R1 rate is being calculated.

According to one of this invention's first application methods, it is expected that a counter will be used to count the requests arising from each source. This will ensure that, during a period, a Il rate renegotiation for each active connection converging on the node in question does take place.
This process is used only for the examination of connections that have the same order number.
According to another of this invention's application methods, it is predicted that the time it takes to change the R1 register will be greater than the network equipment renegotiation period allowed for that purpose.

Appropriately, this period is increased by the maximum time needed by the active connections' RM resource management bits to make the complete trip from the source to the other end of the network and back to the source again. In that way, it is ensured that all sources have the time to comply with the rates authorized by the various network nodes, a compliance which cannot be carried out until the RM resource management bit carrying information on source rate allocation has had the time to return to the source. This way of doing things allows the process to have a certain robustness with respect to the breakdown of certain trunks.

In the case where a connection makes two requests during the same period, the second would not be taken into account so that band sharing between sources is not disturbed.
To do this, each connection is marked once it has made a request. This mark is erased at the beginning of each cycle. Thus, in a given cycle, a request is no longer accepted if it comes from a connection which is already marked.

According to another of this invention's application methods, to avoid analyzing the same connection twice in the same cycle, each cycle is assigned an order number which is increased by one unit in each new cycle. This order number is memorized by each connection after it analyzes the cycle. So let us say that we are in cycle Cy. In this cycle, prior to analysis, a particular connection memorizes the number Cy - 1. After analysis, the number Cy is memorized.

Before allocating a rate to a source, the memorized order number for the connection in question is verified to see if it is one unit lower than that of the current cycle. If this is the case, the allocation is implemented. Otherwise, it is not implemented, since this indicates that the connection in question has already been analyzed in the current cycle.

When cycle Cy is interrupted as a result of a rate request from a non-clipped connection which is greater than that which has been requested to date, and a new cycle is begun, the new cycle will be assigned a number equal to the number of the interrupted Cy cycle plus two units. Thus each verification results in a new allocation.
In the process which has just been described, the rate requested by the video source corresponding to its maximum requested rate is fixed for at least several cycles. The user may modify these parameters if so desired, but in practice, it will rarely be done. If, however, for reasons of network size, or for other considerations, the source requests must be modified very often, we were able to show that the invention's process tends to allocate rates at a value of C/N
with the result that the benefit of being able to use the available band for sharing between the sources whose requirements are the highest is lost.

In the process just described, each source requests a maximum rate. But the process can also be applied in the case where sharing is based on the minimum rates guaranteed from the sources is taken into account. The following procedure accQmplishes this.

The C value of the rate available on the node output link in question is divided by the sum of the minimum guaranteed rate Dmg, (i = 1 to N). A factor is thus obtained which allows the weighting of each equally assigned rate, which we will call R; . We thus have:

R; =Dmg; - C/ I Dmg; =DMG; - C
i=(J...N) _ _ _. ~ . .. _. . . _...~ _.,...~~,,..~...~ ~.~_.,.,..,..-.. _. . _ . _....
.. . . ._ _ _ _ _ .

where DMGi is the weight factor applied to the C; source.

Each Ci connection is thus clipped at the rate of Ri. If we take into consideration the R register in which is stored the total rate assigned to all connections, the R; rate assigned to a connection would equal:
Ri =DMGi =R

At the end of each cycle, the recalculated value of this register will be given by the following relationship which is analogous to the one mentioned above:

C-R= EDMGi +1~
R =R+ i= 1....N =R+C-R+Z~
EDMG; EDMGi unclipped unclipped where Y_DMGi = 1 represents the sum of weight factors, 0 represents the unallocated rate i = 1...,N

and YDMGi represents the sum of weight factors only for connections which have been clipped.
unclipped

Claims (12)

What is claimed is:

1. A method for dynamic rate allocation for implementation in network nodes, each network node having a plurality of sources, each source having a connection to a network node, the method comprising:
allocating to each connection, and in response to one of a plurality of periodic rate requests made by a source corresponding to the connection, a cell rate selected from a lesser of a request rate and a residual rate after allocation of requested rates to all the other sources, wherein, during successive cycles:
a) at the beginning of each cycle of receipt of requests originating from the sources corresponding to the connections, considering and assigning an assigned rate (R1, R i) to each connection; and b) during the cycle, b1) for each connection having a required rate (Rq) which exceeds the assigned rate (R1, R i) for the cycle, allocating the assigned rate (R1, R i) to that and marking it as a clipped connection for the cycle;
b2) for each connection having a required rate (Rq) which is less than the assigned rate (R1, R i), determining whether that connection is marked as a clipped connection from a previous cycle, and if so, allocating the difference between the value of the assigned rate and the value of the rate allocated as a non-allocated rate value (.DELTA.); and if not, allocating the rate already allocated to the connection in previous cycle; and b3) after a rate has been allocated to each connection calculating the number (m) of clipped connections, and calculating a new value for the assigned rate (R1, R i) for the subsequent cycle using the following relation:

wherein C is the rate value of the maximum band offered by the outgoing link at the node, N
is the number of active connections on the node, .DELTA. is the non-allocated rate and m is the number of clipped connections.

2. A method according to claim 1, wherein a rate is assigned to each of the said sources, during an initial cycle, corresponding to equitable sharing among all the sources of the maximum band offered by the outgoing link on the node:

R1 = C/N

where C is the rate value of the maximum band offered by the outgoing link at the said node and N is the number of active connections on the said node.

3. A method according to Claim 1 or Claim 2, wherein each cycle has a period which exceeds a renegotiation period implemented by the equipment on the network.

4. A method according to Claim 3, wherein the renegotiation period is implemented by the maximum duration of the times respectively set by resource management cells on the connections in order to complete the entire path from the source to the other end of the network and return to the source.

5. A method according to any one of Claims 1 to 4, further comprising counting the requests originating from each source and examining only requests with the same serial number.

6. A method according to any one of Claims 1 to 5, further comprising identifying a listing for each connection for which a rate has been requested as a listed requesting connection, examining only the listed requesting connections, and at the end of each cycle, cancelling the listing.

7. A method according to any one of Claims 1 to 6, further comprising weighting each of the rates assigned to the connections using a weighting factor.

8. A method according to Claim 7, wherein each weighting factor is based on a guaranteed minimum rate requested by the corresponding source.

9. A method according to Claim 8, wherein the weighting factor of a source is equal to the guaranteed minimum rate requested by the source divided by the total sum of guaranteed minimum rates respectively requested by all the other sources.

10. A method according to Claim 9, wherein the rate assigned to a connection is equal to the value of a sum of the assigned rates of all the connections multiplied by the weighting factor, and the value of the assigned rate sum is defined by the following relation:

where .DELTA. represents a non-allocated rate and DMG i represents the sum of the weighting factors of the only connections which have been clipped.

11. A method according to any one of the Claims 1 to 10, wherein after receipt of the request from a source whose connection was not marked as a clipped connection in the previous cycle and when the rate now required exceeds the rate which had been allocated to the preceding cycle, the method further comprising interrupting the cycle in progress and re-starting a new cycle from the request, the rate assigned to each source being the greater of the current rate for the connection concerned and the rate value equitably shared among all the sources.

12. A method according Claim 11, wherein each connection retains a number characterising the current cycle.