WO1999027687A1

WO1999027687A1 - Cell dispersion and sequencing method

Info

Publication number: WO1999027687A1
Application number: PCT/FR1998/002459
Authority: WO
Inventors: Yves Peligry; David Mouen-Makoua; Thierry Grenot
Original assignee: Thomson-Csf
Priority date: 1997-11-21
Filing date: 1998-11-18
Publication date: 1999-06-03
Also published as: CA2308367A1; IL135741A0; IL135741A; EP1033014A1; FR2771572A1; FR2771572B1

Abstract

The invention concerns methods for the dispersion and sequencing of cells transmitted by connections (C1....CN1) set up between a source (3i) and one or several destination(s) (4i) connected to intermediate switch modules (2i) defining transmission paths (an). The method consists, for each connection, in considering the different transmission paths (an) as a single transmission resource wherein it disperses the cells. The invention is applicable to any switch, in particular ATM.

Description

PROCESS FOR DISPERSION AND ORDERING OF CELLS

The present invention relates to methods of dispersing and reordering cells. It applies in particular to an ATM switch, the English abbreviation for Asynchronous Transfer

Fashion, to convey digital information of a varied and irregular nature and speed (voice, videos, files).

This digital information forms traffic flows which can be classified into two main categories: - a first category of flow groups real-time traffic. Such traffics are characterized by traffic parameters and time constraint parameters,

- a second category of flow groups traffic of the non-real time type for which the traffic parameters are little or not defined. The transfer of digital information by a switch is accomplished by establishing a connection between an entry point and an exit point of the switch.

In the case of real-time traffic, the necessary resources are reserved according to the traffic parameters. This type of traffic is not a problem.

When it comes to non-real-time traffic, it is difficult to reserve resources based on uncertain parameters. This type of traffic leads to connections with burst activity periods. Under these conditions, the switch must be able to support traffic peaks without loss of cells. The traditional internal routing techniques which consist in reserving a transmission path within the switch according to traffic parameters, and time constraint parameters are ineffective for this type of traffic. In particular, traditional techniques are ineffective for all of the following services, defined by the ATM Forum in specification V.4 entitled "Traffic management":

• ABR service, English initials for Available Bit Rate,

• VBR _NRT service, Anglo-Saxon initials for Variable Bit Rate Non Real Time, • UBR service, Anglo-Saxon initials for Unspecified

Bit Rate. Consider for example a given configuration in which two transmission paths, denoted ai and a2, a portion of which is common, allow each to connect a source to a destination separated by a network of intermediate switching entities. Two connections with sporadic traffic each use one of these two transmission paths. As the connections are sporadic and the number of connections is sometimes large (several thousand), this configuration is very likely. When the two connections transmit at full bandwidth at the same time, congestion occurs on the portion common to ai and a2 regardless of the traffic carried on all the other paths. In this situation, network switching capabilities may be lost; traditional internal routing techniques are ineffective in this case. The under-utilization of the network's switching capacities is increased when part of the traffic corresponds to connections each established between a source and several destinations, the traffic is in this case qualified as "multicast", according to English terminology.

The invention proposes a solution to solve the above problem in an efficient manner, that the traffic corresponds to connections each established between a source and a destination (the traffic is in this case qualified as "unicast", according to English terminology. Saxon) or that the traffic corresponds to connections each established between a source and several destinations.

To this end, the subject of the invention is a method of dispersing and reordering cells transmitted by means of connections established between a source and one or more destination (s) connected by a network of intermediate switching entities defining transmission paths characterized in that it considers all of the transmission paths as a single transmission resource accessible to all connections in which it disperses the cells according to a first determined process and in that it restores the cells in order arrival at destination following a second determined process.

In order to avoid blocking on one of the transmission paths of the network, all of the transmission paths are managed as being a global resource for carrying the traffic of data connections. For this, the cells coming from an entry point, the source, are dispersed in the network to an exit point, the destination in the case of unicast connections or to the exit points, the destinations, in the case of multicast connections. The overall switching capacity of the network is shared by all connections. According to a preferred embodiment of the method, the dispersing process is carried out on the basis of the connection: the cells of the different connections are dispersed on the different transmission paths according to a regular law (or round robin according to English terminology) . The reordering process uses data contained in a connection context attached to the destination.

The invention has the advantage of efficiently managing all the switching capacities of the network and of not losing switching capacities regardless of the type of connection: unicast (a source to a destination) or multicast (a source to several destinations) . The invention will be well understood and other characteristics and advantages will become clear during the following description of particular embodiments, applying both for unicast connections and for multicast connections, illustrated by the appended figures which represent: FIG. 1, an architecture of a switch 1, ATM for example,

FIG. 2, a first connection context,

- Figure 3, a detailed representation of the fields of the first connection context of Figure 2, - Figure 4, a second connection context,

FIGS. 5 and 6, detailed representations of the fields of the second connection context of FIG. 4,

- Figure 7, special cases managed by the reordering process, - Figure 8, different steps of a particular example of reordering cells.

In the various figures, the homologous elements are represented with the same reference.

FIG. 1 represents an architecture of a switch 1, ATM for example. Generally of large capacity, it is based on the use of internal switching modules 2, organized in a network between input modules 3, and output modules 4, Between a given input module 3 and a given output module 4, different transmission paths exist For l example of a switch illustrated in FIG. 1, there are as many different transmission paths as there are intermediate switching modules in the central column, i.e. the transmission paths ai, a2, a3, a4 between the selected modules Between the input module 3ι and the output module ^ N1 connections are to be established, C1,, CN1 with N1 = 100 for example These connections have the characteristic of sporadic cell traffic To efficiently manage the switching capacities of the module network internal switching 2, between input module 3 _! and the output module 4ι, the method considers the different transmission paths ai, a2, a3, a4 as a single transmission resource The cells of the different connections C1,, CN1 are thus dispersed on the different transmission paths ai, a2, a3, a4 To disperse the cells, the method follows a first determined process When they reach their destination, the cells of the different connections C1,, CN1 are put back in order according to a second determined process The second process essentially consists, for a given connection , to store the cells in queues each dedicated to a transmission path, to extract the cells from the queues according to determined criteria to put them in a sequencing queue and to purge the queues when a cell stays there too long

According to a preferred embodiment, the first process and the second process of the method are described below. To disperse and reorder the cells, the method implements different connection contexts. A first connection context, illustrated for information by FIG. 2 is managed by the first process for each connection The first connection context 5 comprises several fields 6, 7, 8 The different fields 6, 7, 8 are detailed in FIG. 3 A first field 6 contains information allowing simultaneously manage connections using the dispersing process and connections not using the dispersion process. Thus, this provision authorizes, for example, the simultaneous management of connections intended to manage in the band intermediate switching entities which do not use the dispersion method with connections supporting normal user traffic which use the dispersion method. According to the illustration, the first field 6 comprises a bit, for example positioned at 1 when it is a connection using the dispersion method. A second field 7 contains a number NEXT_PATH_E indicating the number of the transmission path a _n to be followed by the cell to be dispersed. According to the illustration, the second field 7 comprises three bits making it possible to distinguish a number, NB_PATH, of transmission paths a _n equal to a maximum of eight from a-, to a ₈ , for a given connection. After dispersing a cell in the network, the first process increments the number NEXT_PATH_E for the next cell. When establishing a connection, between the input module 3ι and the output module 4ι, the number NEXT_PATH_E is initiated at a value determined randomly so as to balance the traffic load on the network of entities of switching. A third field 8 contains a number NEXT_SN_E indicating the sequence number of the cell to be dispersed. When establishing a connection, the sequence number NEXT_SN_E is initialized to 0. Before being dispersed, each cell is informed of its sequence number SN = NEXT_SN_E and its transmission path number PATH = NEXT_PATH_E scatter information. These two pieces of dispersion information SN and PATH are stored in a header of the cell independent of the first connection context. A second connection context, illustrated by way of illustration in FIG. 4, is managed by the second process for each connection. The second connection context 9 comprises several fields 10, 11. The different fields 10, 11 are detailed in FIG. 5 and in FIG. 6. A first field 10 detailed in FIG. 5 contains several information. A first piece of information 12 is of the same nature as the information contained in the first field 6 of the first connection context 5. A second piece of information 13 is of the same nature as the information contained in the second field 7 of the first connection context 5. The second information 13 gives the path number of cell transmission expected by the output module for the given connection. Third information 14 is of the same nature as the information contained in the third field 8 of the first connection context 5. The third information 14 gives the sequence number of the cell expected by the output module for the given connection. A fourth piece of information 15 indicates whether or not the second process must take into account a fifth piece of information 16 described below. According to the illustration, the fourth information item 15 comprises a bit, for example set to 1 when the process must take into account the fifth item of information 16 described below. The fifth item of information 16 gives an indication of the time remaining given by a timer, allocated to the second process for reordering the cells. As shown in the illustration, the fifth item of information 16 is coded on 9 bits. A sixth item of information 17 noted NB_DELAYED indicates for a given connection the number of cells waiting, all queues combined. According to the illustration for each of the possible transmission paths (maximum eight because NEXT_PATH_R is coded on three bits) corresponds to a queue, that is to say eight queues per connection. The sixth piece of information 17 comprises ten bits. A second field 11 contains descriptors of the queues attached to the given connection. The second field 11 includes a descriptor Q_DESCRIPT [] per queue. The content of a Q_DESCRIPT [] descriptor is detailed in Figure 6. A Q_DESCRIPT [] descriptor includes several pieces of information. A first piece of information 18 gives the value of a pointer FRONT positioned on the first cell of the queue. As shown in the illustration, the pointer is coded on 12 bits. A second piece of information 19 gives the value of a BACK pointer positioned on the last cell of the queue. As shown in the illustration, the pointer is coded on 12 bits. A third piece of information 20 named FRONT_SN gives the seven least significant bits of the sequence number of the cell which is at the head of the file. A fourth piece of information 21 indicates whether the queue is empty or not. According to the illustration, the fourth item of information 21 comprises a bit, for example positioned at 0 when the queue is empty, at 1 when it is not empty. Taking into account the data previously described, the process diversion is as follows:

When establishing a connection, the process starts with an initialization phase. During this phase, the first process randomly chooses the value of the transmission path NEXT_PATH_E and it initiates the sequence number NEXT_SN_E at zero. Simultaneously the second process initiates the value of the transmission path

NEXT_PATH_R corresponding to the first cell expected at the value chosen randomly by the first process for NEXT_PATH_E and it initiates the sequence number NEXT_SN_R to the same value as the sequence number NEXT_SN_E chosen by the first process, ie zero. Finally, it initiates the descriptors of each queue to zero. Is :

Q_DESCRIPT [0-7] = 0. The queues are considered to be empty, and therefore NBJDELAYED = 0. Outside the initialization phase, the first process and the second process perform different tasks which are described below.

Or a cell to disperse. The header of the cell is filled with dispersion information, ie the sequence number NEXT_SN_E and the number of the transmission path NEXT_PATH_E contained in the first connection context. Which translates into relationships:

SN = NEXT_SN_E (1)

PATH = NEXT_PATH_E (2)

The first process disperses the cell in the network. The cell is sent on the NEXT_PATH_E transmission path. The first process then increments the value of the transmission path modulo the number of possible transmission paths, NB_PATH, and it stores the new value in the first connection context. If the new value of the transmission path NEXT_PATH_E is equal to zero, then the first process increments the sequence number NEXT_SN_E modulo the numbering size of the sequence number and stores it in the first connection context, otherwise it keeps the sequence number NEXT_SN_E unchanged.

When the cell reaches its destination, the second process is activated. It extracts the dispersion information from the cell header: SN and PATH. In a first step, the second process checks whether the received cell is inserted in the reception window. A first end of the reception window is determined by the waiting cell identified by the values of the second connection context NEXT_SN_R and NEXT_PATH_R. A second end is located at a distance from the first end equal to the size of a queue, 128 cells for the example considered. Verification can be performed by testing the following expression:

X = Unsigned value of [SN - NEXT_SN_R]

IF (X = 0 and the number queue NEXT_PATH_R is not empty)

THEN the cell received is not in the reception window

OTHERWISE IF X <128

THEN the received cell is in the reception window ELSE the received cell is not in the reception window.

Depending on the test result, several cases should be considered for the course of the second process:

1st case: the received cell is in the reception window. Two sub-cases are to be distinguished: 1st sub-case: the received cell corresponds to the expected cell.

In other words, on the one hand

PATH = NEXT_PATH_R (3) and on the other hand

SN = NEXT_SN_R (4) Under these conditions, the received cell is stored in the sequencing queue. The NEXT_PATH_R and NEXT_SN_R information is updated according to the PATH and SN values of the cell. Which translates into relationships:

NEXT_PATH_R = PATH + 1 (modulo NB_PATH) (5) IF NEXT_PATH_R ≠ 0

THEN NEXT_ SN_R - SN (6)

OTHERWISE NEXT_ SN_R = SN + 1 (modulo the numbering size of SN). (7)

The TIMEOUT_DATE timer is inhibited, by setting the TV bit to O. 2nd sub-case: the cell received does not correspond to the cell expected. If the timer is not activated (TV = 0), then the timeout end date TIMEOUT_DATE is calculated and the indication is given that the timer is active by setting TV to 1. The cell received is stored in the queue d 'wait corresponding to the transmission path it followed, ie in the PATH number queue. The number of cells waiting for the NB_DELAYED connection is incremented by 1. If the queue of PATH number was empty, then it is declared not empty, the fourth information 21 of the descriptor Q_DESCRIPT [] of the queue is positioned at non-empty, S = 1. The 7 least significant bits of the field SN extracted from the cell are reported in the field FRONT_SN of the descriptor of the queue.

2nd case: the cell received is outside the reception window. A transmission incident took place and many cells were lost. Under these conditions, the queues corresponding to the connection are emptied, the number of cells waiting for the connection is set to zero NB_DELAYED = 0, the time delay is inhibited, this is obtained by initializing TV at 0 and the cell received is placed in the sequencing queue. The NEXT_PATH_R and NEXT_SN_R information of the second connection context is updated according to the PATH and SN values of the cell.

Which translates into relationships:

NEXT_PATH_R = PATH + 1 (modulo NB_PATH) (5)

IF NEXT_PATH_R ≠ 0 THEN NEXT_SN_R = SN (6)

OTHERWISE NEXT_SN_R = SN + 1 (modulo the numbering size of SN). (7)

In addition to receiving cells, the second process extracts cells from the queues and puts them in the sequencing queue. For a given connection, following the reception of a cell, or when a cell of this connection has been extracted from the sequencing queue, periodically, the second process traverses the queues. If they are empty, the second process does nothing. If they are not empty, several cases may arise. Figure 7 shows four special cases A, B, C, D. In the different cases, the second process is waiting for the cell of sequence number equal to 2 (NEXT_SN_R = 2) on the transmission path equal to 1 (NEXT_PATH_R = 1). For each case, only four queues F0, F1, F2, F3 were represented. Case A: the expected cell has not arrived; timing

TIMEOUT_DATE is initialized if it was not already.

Case B: the expected cell SN = 2 is present in the expected queue PATH = 1. The cell is extracted from the queue F1 and stored in the sequencing queue. The NEXT_PATH_R and NEXT_SN_R information of the second connection context is updated and the number of cells waiting for the NB_DELAYED connection is decremented by 1.

Case C: the expected cell has not arrived but the queue F1 is not empty, which means that at least one cell has been lost on the corresponding transmission path. The second process must then find the next cell to store in the sequencing queue. In the example two cells were lost, the cell SN = 2 on the transmission path equal to 1 and the cell SN = 2 on the transmission path equal to 2. Indeed, the dispersion of the cells on the different paths of transmission follows a regular law: the cells are distributed regularly from the transmission path equal to 0, which corresponds to the queue F0, to the transmission path equal to n, which corresponds to the queue Fn. The next expected cell is thus the cell with a sequence number equal to 2 on the transmission path equal to 3. This cell is extracted from the queue F3 and placed in the sequencing queue. The NEXT_PATH_R and NEXT_SN_R information is updated and the number of cells waiting for the NB_DELAYED connection is decremented by 1.

Case D: the conditions are the same as in case C except that the queue F3 is empty. The timeout END TIMEOUT_DATE date is calculated, the timeout is activated by setting TV to 1, if it was not already and the values of NEXT_PATH_R and NEXT_SN_R of the second connection context are updated. That is, NEXT PATH R = 3 and NEXT SN R = 2. Whatever the case encountered, the extraction of a cell follows the following algorithm:

- for each queue an SNF sequence number is determined: 1st case: the queue is not empty; in this case the SNF sequence number is determined from the third piece of information 20 of the descriptor Q_DESCRIPT [] of the queue. Thus SNF = FRONT_SN

2nd case: the queue is empty; in this case a virtual value is calculated as follows: either NFILE the number of the queue IF NFILE <NEXT_PATH_R

THEN SNF = NEXT_SN_R + 1 OTHERWISE SNF = NEXT_SN_R - for each queue a distance between the cell with sequence number SNF and the expected cell is calculated. This distance is given by the following expression: distance (NFILE) = (NB_PATH * [SNF-NEXT_SN_R] + [NFILE-NEXT_PATH_R]) (modulo NB_PATH * 128) - the queue selected is the one whose distance, distance (NFILE), is minimal.

- if the selected queue is empty, the TIMEOUT_DATE timer is activated if it was not already. If the selected queue is not empty, then the first cell of the queue is extracted and stored in the sequencing queue. The information NEXT_PATH_R and NEXT_SN_R of the second connection context are updated and the number of cells waiting for the connection NB_DELAYED is decremented by 1. Thus, taking again case A of figure 7, it is possible to calculate the distance, distance (NFILE), for each queue F0 to F3: for F0 distance (0) = 4 x (3-2) + (0-1) = 3 for F1 distance (1) = 4 x (2-2) + (1 -1) = 0 for F2 distance (2) = 4 x (2-2) + (2-1) = 1 for F3 distance (3) = 4 x (2-2) + (3-1) = 2 The queue selected is the F1 queue because it corresponds to the minimum distance. The queue F1 is empty, therefore the TIMEOUT_DATE timer is activated if it was not already. For case C, the algorithm leads to the following distance values: for FO distance (0) = 4 x (3-2) + (0-1) = 3 for F1 distance (1) = 4 x (3-2 ) + (1-1) = 4 for F2 distance (2) = 4 x (3-2) + (2-1) = 5 for F3 distance (3) = 4 x (2-2) + (3-1 ) = 2

The queue selected is queue F3 because it corresponds to the minimum distance. The queue F3 is not empty, the cell SN = 2 is stored in the sequencing queue. The NEXT_PATH_R and NEXT_SN_R information of the second connection context is updated: NEXT_PATH_R = 0 and NEXT_SN_R = 3, considering that the number of NB_PATH transmission paths is equal to 4. The number of cells waiting for the NB_DELAYED connection is decremented from 1.

The TIMEOUT_DATE delay common to all the different transmission paths of a connection makes it possible to limit the waiting times of an expected cell. When the timer expires, the queues are emptied, the number of cells waiting for the connection is set to zero NB_DELAYED = 0. the timer is inhibited by setting TV to 0 and the first process and the second process are reset. FIG. 8 gives examples of reordering cells in a switch in which the cells have been dispersed over four transmission paths to which correspond four queues F0, F1, F2, F3. Seven successive steps are illustrated.

Step 1: no cell is present in the different queues, F0 to F3. The expected cell is the cell with the sequence number SN = 1 on the transmission path 1 corresponding to the queue F1.

Step 2: a cell coming from the transmission path 2 is received with the sequence number SN = 1. It is not the expected cell, it is therefore stored in the queue F2. The end date of TIMEOUTJDATE timer is calculated, the timer is activated by setting TV to

Step 3: new cells are received, but since the expected cell (SN = 1 in F1) has still not been received, the cells remain stored in the queue corresponding to the transmission path followed by the cell.

Step 4: the expected cell has been received, it has been placed in the FS sequencing queue. The NEXT_SN_R and NEXT_PATH_R information of the second connection context is updated: NEXT_PATH_R is incremented either NEXT_PATH_R = 2 and NEXT_SN_R is unchanged, or NEXT_SN_R = 1.. The number of cells waiting for the NB_DELAYED connection is decremented by 1.

Step 5: the expected cell (SN = 1 in F2) is present, it is extracted and stored in the sequencing queue FS. No cells were received. The information NEXT_SN_R and NEXT_PATH_R of the second connection context are updated: NEXT_PATH_R is incremented either NEXT_PATH_R = 3 and NEXT_SN_R is unchanged, ie NEXT_SN_R = 1. The number of cells waiting for connection NB_DELAYED is decremented by 1. Step: 6 the expected cell (SN = 1 in F3) is present, it is extracted (NB_DELAYED = NB_DELAYED-1) and stored in the sequencing queue FS. The cell SN = 2 is received on the transmission path equal to 3. It is stored in the queue F3 (NB_DELAYED = NB_DELAYED + 1). The NEXT_SN_R and NEXT_PATH_R information of the second connection context is updated: NEXT_PATH_R is incremented either NEXT_PATH_R = 0 (because the modulo is 4) and NEXT_SN_R is incremented because NEXT_PATH_R = 0, or NEXT_SN_R = 2.

Step 7: the expected cell (SN = 2 in FO) is present, it is extracted (NB_DELAYED = NB_DELAYED-1) and stored in the sequencing queue FS. The cell SN = 2 is received on the transmission path 1. It is stored in the queue F1 (NB_DELAYED = NB_DELAYED + 1). The NEXT_SN_R and NEXT_PATH_R information of the second connection context is updated: NEXT_PATH_R is incremented either NEXT_PATH_R = 1 and NEXT_SN_R is unchanged, or NEXT_SN_R = 2.

Claims

1. Method for dispersing and reordering cells transmitted by means of connections established between a source (3ι) and one or more destination (s) (4 |) connected by a network of intermediate switching entities (2j) defining transmission paths (a _π ) characterized in that it considers all of the transmission paths (a _n ) as a single transmission resource accessible to all connections (C1, ..., CN1) in which it disperses the cells according to a first determined process and in that it restores the cells arrived at destination (4d) according to a second determined process.

2. A method of dispersing and reordering cells according to claim 1, characterized in that for a given connection the first process consists:

- to manage a first connection context (5) attached to the source 3) and comprising a first (NEXT_SN_E) and a second (NEXT_PATH_E) dispersion information updated after each transmission from a cell to distribute the cells over the different transmission paths (a1, a2, a3, a4) according to a regular law,

- to inform each cell with the current dispersion information (SN, PATH) before transmitting it on the transmission path corresponding to the second dispersion value (PATH) contained in the cell, and the second process consists for a destination (4 -,) given:

- extracting the dispersion information from the header of each cell received (SN, PATH) and managing a second connection context (9) attached to the destination (4 to decide to store the cell received in a queue determined (FO, F1, F2, F3) or put it in a sequencing queue (FS), to decide to purge the queues (FO, F1, F2, F3) or to decide to extract a cell from '' a queue (FO, F1, F2, F3) determined to store it in the sequencing queue (FS).

3. A method of dispersing and reordering cells according to claim 2, characterized in that the management of the first connection context (5) attached to the source (3 ^ consists of initializing, when establishing the connection or during a reset, the first dispersion information (NEXT_SN_E) to a given value and the second dispersion information (NEXT_PATH_E) to a value determined randomly then to increment, after each cell emission, the first dispersion information (NEXT_SN_E ) modulo a first determined value and the second information of dispersion (NEXT_PATH_E) according to a regular law and modulo a second determined value and in that the information of the cell consists in copying in a header of the cell the current values of information of dispersion (NEXT_SN_E, NEXT_PATH_E) contained in the first connection context (5).

4. A method of dispersing and reordering cells according to claim 3, characterized in that the management of the second connection context (9) consists of:

- to identify a cell expected by a sequence number (NEXT_SN_R) and a transmission path number (NEXT_PATH_R),

- to initialize, when establishing the connection or during a reinitialization, the sequence number (NEXT_SN_R) to the initialization value of the first dispersion information (NEXT_SN_E) and the transmission path number (NEXT_PATH_R ) the initialization value of the second dispersion information (NEXT_PATH_E),

- to update the sequence number (NEXT_SN_R) and the transmission path number (NEXT_PATH_R) of the expected cell after each cell reception,

- to define a reception window according to the expected cell,

- detecting a transmission problem when the received cell arrives outside the reception window,

- purge the queues (F0, F1, F2, F3) if a transmission problem has been detected, - to compare, if no transmission problem has been detected, the received cell with the expected cell to decide on the ordering of the received cell in a queue (F0, F1, F2, F3) or in the sequencing queue (FS),

- to allocate a maximum waiting time for waiting for an expected cell beyond which the queues (F0, F1, F2, F3) are emptied and the first and second processes are reset.