DE102012207380A1 - Method for distributing data streams in a network element - Google Patents

Abstract

The invention relates to a method for distributing data streams having data cells assigned to a time segment in a network element for connecting network segments, the network element having a plurality of outputs and inputs, comprising the following steps: analyzing at least a first and a second data stream, each abutting a first and a second input; Determining if a first contiguous data cell from the first data stream in the same time period is addressing the same output as a second contiguous data cell from the second data stream; Determining an unused input of the network element; Redirecting one of the competing data cells to the unused input. The invention further relates to a corresponding control device and a corresponding network element.

Description

The invention relates to a method for distributing data streams in a network element, a control unit for a network element and a network element for connecting network segments

A network switch is a coupling element that connects network segments. For example, several PCs or other devices are connected to a WAN (Wide Area Network) via a switch. For this purpose, a switch has so-called "input ports" or data inputs or inputs at which data streams are created as well as "output ports" or data outputs or outputs on which data streams leave the switch again.

It may now be the case that individual data cells from data streams at different inputs address the same output at the same time. In the above example, this would occur if e.g. two or more PCs or other devices want to access the same network resource. As a result, the two data cells from different input data streams must share the same output. Depending on the structure of the incoming input data stream delays may occur and in particular even with data cells that address free outputs, but are blocked by "not fetched" cells or such cells that address the same output and whose forwarding is therefore delayed. This problem is also known as "Head of Line Blocking".

Depending on the size of an input memory associated with an input, during this delay the input memory may fill up with additional data cells, so data must be discarded. In so-called FIFO ("first in first out") input memories, in this case the data first arrived are removed from the memory and thus lost.

Often referred to as so-called "Core Switches" high-quality high-performance switches, which are equipped with a complex organized, larger buffer memory for the "I / O" output or input / output Ausgabge, so that data loss due to blocking effects can be avoided. However, these solutions are associated with high costs and are therefore used at critical points of a computer network, for example in the "core" or core area of a network or in the so-called "backbone" network, where usually very high data transmission rates occur.

In non-central network areas, which are often connected to groups of terminal devices, often referred to as workgroups or "workgroup" areas, high-quality, high-performance core switches are not used for cost reasons. Rather located in the low price segment so-called "work group switches", which find their application here, have no such complex buffer memory. These work group switches usually implement the so-called "input queuing", i. an input has an input buffer memory. As long as the data streams now follow a uniform distribution between the outputs, these switches are sufficiently powerful. However, if the distribution of data streams at the inputs changes, blockages can occur that significantly affect the switching performance of the switch.

It is therefore an object of the present invention to provide a way to connect network segments easily and inexpensively.

This object is achieved by a method, a control unit for and a network element according to the independent claims. Further advantageous embodiments of the invention are the subject of the dependent claims.

According to the invention, it is determined whether there are competing data cells in data streams which apply to respectively one input of a network element or switch, i. in more than one data stream at a particular time period the same output of the network element is addressed. If this is the case, one of the competing data cells is redirected to an unoccupied input, the other, remaining data cell remains at the respective input.

This has the advantage that data cells located behind the redirected data cell, which address another output, are no longer blocked or their forwarding to the other output is not unnecessarily delayed by the redirected data cell. This ensures a higher throughput through the network element and possibly avoids data loss that occurs when a buffer belonging to the input has become full due to the delay. Advantageously, the solution according to the invention can be used in the workgroup and / or in the core area.

In particular, the remaining data cell and the redirected data cell are then distributed from different inputs to the output using known methods. This ensures a rapid forwarding of competing data cells.

According to an advantageous embodiment, for the uniform distribution of data cells at different inputs to the same output or the processing of input queues at different outputs, methods according to the "fair queuing" and / or the "round robin" and / or the "priority queuing" are used ,

In particular, the structure of the data streams is examined, for example with regard to the addressing information of the individual data cells. Based on this, the decision is made which of the competing data cells will be redirected.

According to an advantageous embodiment, in the case of a burst-like data stream, i. a data stream in which the individual data cells repeatedly address the same output, no diversion of a data cell from this bursty data stream but the competing data cell from the other data stream.

This has the advantage that, in this case, it is not necessary to redirect the subsequent data cell at the next data cell, or in a next work cycle of the network element, but instead block the blockade, for example. in the case that in the other data stream only one data cell is blocked or the same output is addressed, it can be canceled.

In particular, the redirection of the redirected data cell by forwarding to the output, which is assigned to the unused input and is directly connected to this or short-circuited.

This has the advantage that virtually an input data stream is applied to this input or that a virtual input queue of this input can be used as a buffer memory extension and thus - without hardware changes to the network element itself - unused resources in the network element can be used.

In particular, the determination as to whether an input is unused is made by whether something is connected to that input, that input is turned on or de-energized, is in a stand-by function, or otherwise does not receive data for a sufficient amount of time. These situations can be easily detected and evaluated via control signals.

In particular, even if a data cell from a data stream to another input in turn addressed the same output, this also redirected to a free input.

In particular, if several outputs are blocked or address at least two data cells at the same time at least two outputs, one free input is determined in each case and there is a respective diversion, so that the blockade is resolved.

According to an advantageous embodiment, if more than two data cells address the same output, the blockage is successively resolved, i. First, one of the competing cells is redirected, then in a subsequent work cycle another one of the competing cells, and so on. Alternatively, according to another embodiment, a redirection takes place in the same work cycle.

In particular, the competing data cells are determined for a current period of time, in particular at the head of the input queue. Thus, blockages can be resolved as they occur.

Alternatively or additionally, competing data cells are determined for periods of time lying in the future. Thus, a plan can be made on which free inputs should be redirected to ensure a permanently high data throughput.

The invention also relates to a control unit for a network element, which is set up to redirect data cells in the network element according to the method according to the invention.

The invention further relates to a network element with such a control unit. This allows the implementation of the method in a compact device.

Alternatively, the network element has an interface for communication with such a control unit. This has the particular advantage that a plurality of network elements are controlled by a control unit. This has cost advantages and further enables the coordination of individual network elements to use free resources even more effectively.

Further advantages of the invention will be explained by means of figures, of which show:

1 a prior art queuing architecture that uses one input queue per input;

2 a blocking state in which a data cell C21 at input 2 blocks a subsequent data cell C22;

3 a representation of a blockage in which two data streams at the same time a data cell occurs, which addresses the same output;

4 the first cycle of a resolution of in 3 shown blocking;

5 the second cycle of the resolution of the blocking, which in 3 is shown, and

6 the third cycle of the dissolution of the 3 shown blocking.

In 1 a switch or network element S is shown, which has a plurality of input ports or inputs E1, E2, ... En. Each of the inputs E1, E2,... En has data streams which can be subdivided into data cells D which have a fixed length of time. Via a corresponding interconnection, the data cells D are routed to an output A1, A2,... An, which is addressed by the data cell D in each case.

One or more data cells are processed in a work cycle of the network element operating at a clock rate.

According to an advantageous embodiment, a data cell D with a data cell-based switch architecture is a fixed-length data unit in a data stream applied to an x-th input Ex or in an x-th input queue IQx, where x is a number {1, 2, ..., n} is.

The fixed length is usually specified over a number of bytes or byte number. By means of this byte number, the defined time length or time span or time duration is defined via the clock frequency with which the network element S operates. In turn, the period of time is defined in which a forwarding of the data cell to an output takes place on schedule, so that it can be determined which data cells from different input data streams lie in the same time segment.

According to a further embodiment, which uses a frame or "frame" based switch architecture, a data cell represents a frame, ie a data unit with variable length or number of bytes in a data stream lying on an x-th input Exan or in an x -th input queue IQx of fixed length, where x is a number of {1, 2, ..., n}. Again, can be found on the clock frequency, a period of time occupied by the frame.

At each of the inputs E1, E2,... En is a buffer memory PS in which a predetermined number of data cells D can be stored. As long as the distribution of the data streams from the individual inputs E1, E2,... En to the respectively addressed outputs A1, A2,... An is uniform, the distribution according to FIG 1 a sufficient data throughput through the network element S.

However, this concept of simple "input queuing" has the disadvantage that as soon as an addressed output A1, A2, ... An or output port is not free, all other data cells D in the respective input queue or such cells D, which are stored in the respective buffer PS of the relevant input E1, E2, ... En, although they may eventually address a free output. This blocking represents a significant impairment.

A corresponding blocking condition is in 2 shown. At the input E1 are the data cells D, which are marked with C11 and C11, in the buffer PS or form the input queue or "input queue" at the input E1. These data cells D address the output A1. At the input E2 are the data cells D, which are marked with C22, C21, in the input queue or are stored in the input memory PS from input E2. The data cell D, which is marked C21, also addresses the output A1.

This leads to delays in the forwarding of the subsequent data cell D, which is marked C22, although C22 addresses the free output A2. As nomenclature for the data cells D, it has been chosen that the cell Cij is located in the input queue i and addresses the output j.

To eliminate this blocking state, it is provided that a control unit STE analyzes the data streams arriving at the individual inputs E1, E2,... En to determine whether a blocking state is present or will occur in the future.

The control unit STE is according to 2 integrated in the network element S.

Alternatively (not shown), the network element S has an interface over which data is received from an external control unit. This has the advantage that several network elements S are controlled by a control unit, so that on the one hand only a control unit is necessary, which further costs can be reduced and on the other coordination measures between individual network elements S can be made. In this embodiment, therefore, not every network element necessarily has a control unit STE.

• – Vorliegen von burst-artigen Datenströmen, d.h. ein Datenstrom adressiert immer wieder denselben Ausgang, beispielsweise setzt sich die Eingangswarteschlange des Eingangs IQi aus Datenzellen Cij, Cij, Cij, Cij, ... zusammen. Dies ist in 3 dargestellt.
• – Eine weitere Eingangswarteschlange IQm adressiert jeweils freie Ausgänge.
• – In der Eingangswarteschlange des Eingangs El, also der Eingangswarteschlange IQl adressiert die Datenzelle D, welche mit Clj gekennzeichnet ist, zur selben Zeit bzw. im selben Arbeitszyklus, den gleichen Ausgang Aj wie die Datenzelle D Eingangswarteschlange IQi, welche mit Cij gekennzeichnet ist. Dadurch wird die Weiterleitung von nachfolgenden Datenzellen D in der Eingangswarteschlange IQl, z.B. der mit Cle gekennzeichneten Datenzelle verzögert bzw. diese Datenzelle wird blockiert, obwohl die mit Cle gekennzeichnete Datenzelle den freien Ausgang Ae adressiert.

In the context of this analysis, according to one embodiment, the addressing information of the data streams or input queues present at the respective inputs or the respective contents of the buffer memories PS of the individual inputs E1, E2,... En are monitored and evaluated as follows:

• Presence of burst-like data streams, ie a data stream repeatedly addresses the same output, for example the input queue of the input IQi is composed of data cells Cij, Cij, Cij, Cij,... This is in 3 shown.
• - Another input queue IQm addresses each free outputs.
• In the input queue of the input El, that is to say the input queue IQ1, the data cell D marked Clj addresses the same output Aj as the data cell D input queue IQi marked Cij at the same time or duty cycle. This delays the forwarding of subsequent data cells D in the input queue IQ1, eg the data cell marked Cle, or blocks this data cell, although the data cell labeled Cle addresses the free output Ae.

The following figures describe the resolution of this blockade state:

For this purpose, the control unit STE or the control logic is looking for an input which is not occupied, to which, for example, no PC is connected. This is often the case, as input / input port reserves are often planned and installed for the so-called workgroup switches in order to expand the network at a later time if necessary. In operation, therefore, often not all inputs E1, E2, ... En occupied simultaneously.

It can also happen that not all terminals at the occupied inputs E1, E2, ... En are switched on at the same time.

Furthermore, within the scope of so-called "Green IT" solutions, temporarily unused inputs may be switched to standby for communication. This has the consequence that in the network element S unused resources such as the input queues or input memory PS are present.

Via a suitable control unit STE, the states of the input ports "not occupied", "terminal not switched on" and "input switched to standby" can be determined.

Thus, the control unit STE determines an input that is currently not needed. In 4 this free input is represented as input En. The output An is switched directly with the associated input En, ie it is short-circuited. The data cell D, which in 3 is marked with Clj, is routed via the output An to the unoccupied input queue IQn. In the subsequent processing cycle, the data cell D, which in 3 marked with Cle, will be forwarded. The input queue IQn is therefore used as a virtual buffer extension. The prioritization of the forwarding of data cells D from different input queues IQ or of data streams applied to different inputs is carried out by means of known methods, for example the so-called "fair queuing".

Alternatively or additionally, distribution methods such as "round robin", "priority queuing", etc. are used.

The relevant distribution method is configured in the network element S. When one or more unused inputs Ex are activated, they receive the input queue configuration of the network element applied analogously, derived from the input queue using the unused input.

In the following, blocking data cells D are redirected until the first data cell in an input queue IQ addresses a free output A. In 5 It is now shown how the data cell D at the input En, which is marked Cij, is routed to the output Aj. The data cell located at the input Ei, which is marked Cij, still has to wait, ie its forwarding is delayed. The data cell D, which is located at the input El and marked Cle, is routed to the output Ae.

In the following switching cycle, the in 6 is shown, the remaining data cell D at the input Ei, which is marked with Cij, forwarded to the output Aj.

If blocking or delay of the forwarding concern several outputs A, so according to an advantageous embodiment per output A, a free input E whose input queue IQ is used, searched.

According to an advantageous embodiment, data cells in burst-like data streams, ie when the same output A is addressed for each data cell D, not redirected in the case of a blocking state, but a data cell D in the other input queue IQ redirected. This has the background that in the case of the burst-like data streams, which, for example, in 3 with Cij also address the same output Aj and thus again there is a blockade. It is therefore better to redirect a data cell D located in a buffer PS of an input E to which no burst-like data stream is applied.

According to an advantageous embodiment, the control unit STE is formed by a microcontroller, which is programmed by means of an appropriate software. Alternatively, the control unit STE is formed by a specifically manufactured ASIC.

In particular, in further embodiments, the order of the method steps can be varied, so that, for example, the determination of unused inputs takes place only at predetermined time intervals, while in each execution cycle the presence of competing data cells is determined. Furthermore, the determination of the contention situation for the current execution cycle or / and for a future execution cycle can take place. In particular, the aspects of the different embodiments can be varied and / or combined with one another.

LIST OF REFERENCE NUMBERS

• A

  output

  D

  data cell

  e

  entrance

  B

  blockade

  NB

  non-blocking condition

  IQI

  Input queue at input I

  STE

  control unit

  PS

  buffer memory

Claims (11)

A method of distributing data streams having data cells associated with a period of time in a network element for connecting network segments, the network element having a plurality of outputs and inputs, comprising the steps of: a. Analyzing at least a first and a second data stream respectively applied to a first and a second input; b. Determining if a first contiguous data cell from the first data stream in the same period of time addresses the same output as a second contiguous data cell from the second data stream; c. Determining an unused input of the network element; d. Redirecting one of the competing data cells to the unused input. Method according to Claim 1, with the following further step: e. Forwarding the contending data cell and the data cell to the addressed input for a method of distributing data cells at different inputs to different outputs. The method of claim 1 or 2, wherein the structure of the data streams is analyzed and the decision of which of the contending data cells is redirected is made depending on the structure of the data streams. Method according to one of the preceding claims, in which the diversion is realized by forwarding to the output belonging to the unused input and this output is short-circuited with its input. Method according to one of the preceding claims, wherein for determining whether an input is unused, it is examined whether a connection is connected to the input and / or the input is switched to "de-energized" and / or to a "standby" function is switched and / or no data stream is present at the input over a specified period of time. Method according to one of the preceding claims, in which at least one further competing data cell is detected in at least one further data stream and this is redirected to at least one further unoccupied input. Method according to one of the preceding claims, in which a free input is determined for each output which is addressed simultaneously by more than two data cells. Method according to one of the preceding claims, in which the time period relates to a current or / and a future working cycle. Control unit for a network element with an interface for communication with at least one network element, wherein inputs of the at least one network element can be connected to outputs by means of the control unit, and the control device is set up to carry out a method according to one of the preceding claims. A network element for connecting at least two network segments to - a control unit according to claim 9 - a plurality of inputs to each of which a data stream is applied, which is temporally divided into data cells; - A plurality of outputs, wherein an output is connectable by means of the control device with an input. Network element for connecting at least two network segments with - An interface to a control unit according to claim 9 - A plurality of inputs to each of which a data stream is applied, which is divided into temporal data cells; - A plurality of outputs, wherein an output is connectable by means of the control device with an input.

Images