JP5359692B2 - Frame output method and frame output device - Google Patents

Description

  The present invention relates to a frame output method and a frame output device for reading out and outputting a frame from a frame buffer.

  In recent years, high-speed and large-capacity layer 2 switch (hereinafter referred to as “L2 switch”) devices are progressing, and accordingly, a higher transmission rate per port and a higher number of ports per device are required. ing. As the speed and capacity increase, communication quality QoS (Quality of Service) processing needs to be performed at high speed, which also affects the scheduler that is an egress (output side) function of the L2 switching device.

  FIG. 1 shows a configuration diagram of an example of an L2 switch. In FIG. 1, the frames input from the ports # 1 to #M are assigned to the queues # 1 to #N of the ingress unit 1 and the input rate is adjusted based on the header information. A frame output from the ingress unit 1 and reaches the egress unit 3 through the switching unit 2 is stored in the destination queues # 1 to #N of the frame buffer based on the header information.

  The scheduler, which is a function of the egress unit 3, determines a port and a destination queue from which frames are read out by round robin based on frame accumulation information and token presence / absence information (and destination queue priority) for each destination queue. The scheduler reads the frame in page units from the determined destination queue in the frame buffer divided into page units of a predetermined length by following the pointer chain. Then, the token corresponding to the read frame length is subtracted. Frames can be read from the queue until there are no more tokens, and a fixed amount of tokens is periodically added. Frames read from the frame buffer are output from ports # 1 to #M.

  FIG. 2 shows a configuration diagram of an example of a conventional egress section. In FIG. 2, the frame is input from the switching unit 2 and supplied to the header processing unit 11. The header processing unit 11 adds an in-device header to the head of the frame as shown in FIG. The in-device header includes destination queue information of the frame output destination and length information (frame length) indicating the length of the frame. The frame with the in-device header is supplied to the buffer control unit 12, and the destination queue information and the frame length extracted from the in-device header are supplied to the pointer management unit 13.

  The write pointer issuing unit 13b of the pointer management unit 13 accesses the tail pointer management memory 13a and the chain management memory 14, issues a write pointer to the queue indicated by the destination queue information, and sends it to the write control unit 12a of the buffer control unit 12. Supply.

  The tail pointer management memory 13a manages the last pointer of the queue for each queue, and the chain management memory 14 manages the queue pointer chain and frame length for each queue. As a result, the write control unit 12a writes the frame in the frame buffer (memory) 15 using the write pointer as an address.

  The tail pointer management memory 13a supplies enqueue information (frame input information) to the frame presence / absence information unit 16a of the scheduler unit 16 by issuing a write pointer. As a result, the frame presence / absence information unit 16a recognizes that there is a frame in the queue and notifies the round robin unit 16b in the scheduler unit 16 of it.

  The token counter 16c of the scheduler unit 16 counts tokens for each queue. A fixed amount of tokens is periodically added, and the frame length is subtracted every time a frame is read from the queue. Note that the token presence / absence information for the queue whose count value is zero or minus indicates that there is no token.

  The round robin unit 16b is notified of token presence / absence information from the token counter 16c, and is notified of frame presence / absence information from the frame presence / absence information unit 16a. The round robin unit 16b determines a queue to be read by the round robin method. The round robin unit 16b determines an entry candidate when there are both a token and a frame, and determines a queue to be read from the entry candidates according to the priority of the queue. Then, the round robin unit 16b supplies the queue information determined to be a read request to the read pointer issuing unit 13c of the pointer management unit 13. At this time, if there is no token or frame, no read request is output.

  The read pointer issuing unit 13c accesses the head pointer management memory 13d that manages the head pointer of the queue with the determined queue information and the chain management memory 14 to read the page unit by reading the pointer chain. Issue pointers sequentially.

  Then, the read pointer issuing unit 13c supplies the issued read pointer as an address to the read control unit 12b of the buffer control unit 12. The read control unit 12b reads a frame from the frame buffer 15 in units of pages.

  The read pointer issuing unit 13c subtracts the frame length of the frame stored in the chain management memory 14 from the value of the token counter 16c when the read pointer is issued.

  The head pointer management memory 13d stores the frame presence / absence information of the queue in the frame presence / absence information unit 16a when the next frame (next chain) in the queue (queue from which the frame has been read) stored in the chain management memory 14 is exhausted. Update with no frame. As a result, the frame presence / absence information unit 16a recognizes that there is no frame in the queue and notifies the round robin unit 16b in the scheduler unit 16 of it.

  The read frame is temporarily stored in a buffer for each port in the output buffer unit 17 and output to the outside.

  Conventionally, proposals have been made for a large-capacity frame switch and a scheduling function that realize complex control such as QoS control and discard control at a high speed and with a small hardware scale for variable-length frames (see, for example, Patent Document 1). ).

JP 2001-339427 A

  In order to guarantee the set output bandwidth, it is necessary to perform the time from the determination of the frame to be read by the scheduler to the determination of the next frame to be read at a certain period.

  When determining a queue to be read by the round robin method, the scheduler unit 16 sets a round robin entry candidate when both a token and a frame exist, and does not output a read request when there is no token or frame. At this time, since the frame presence / absence information and the token presence / absence information are not determined until after the previous frame is read, the time T1 for performing the round robin processing as shown in FIG. Only a short time is allocated after the frame presence / absence information and token presence / absence information are determined.

  In FIG. 4, since there is a frame and a token at time t1, a frame read request is output at time t2. Also, since there is no frame and no token at time t3, no frame read request is output at time t4.

  The scheduler unit 16 will break the transmission rate unless the time from when the frame presence / absence information and the token presence / absence information is determined until the read request is output within a certain time. Therefore, the exchange of information between the pointer management unit 13 and the scheduler unit 16 must be performed at high speed. However, when the number of queues increases, the time required for the scheduler unit 16 to search for each queue and determine a queue to be read increases, and there is a problem that the circuit design of the scheduler unit 16 breaks down.

  An object of the disclosed frame output apparatus is to perform the next scheduling in advance without waiting for completion of frame reading.

A frame output device according to an embodiment of the disclosure generates an address for writing an input frame to a frame buffer for each queue according to a destination, and generates a pointer management unit that generates an address for reading the frame from the frame buffer;
Scheduler means for adjusting a frame reading order from the frame buffer and generating a frame reading request for each queue;
Request buffer means for accumulating frame read requests generated by the scheduler means, sequentially taking out the stored frame read requests and supplying them to the pointer management means ;
An accumulated number counting means for counting the number of accumulated frames for each queue of the frame buffer and transmitting it to the scheduler means;
The frame read from the frame buffer is output at the address generated by the pointer management means.

  According to the present embodiment, the next scheduling can be performed in advance without waiting for completion of frame reading.

It is a block diagram of an example of L2 switch. It is a block diagram of an example of the conventional egress part. It is a figure which shows the flame | frame which added the header in a device. 3 is a signal timing chart of each part of the circuit in FIG. 2. It is a block diagram of one Embodiment of an egress part. It is a figure which shows the header in an apparatus. 6 is a signal timing chart of each part of the circuit in FIG. 5.

  Embodiments will be described below with reference to the drawings.

<Configuration of egress section>
FIG. 5 is a configuration diagram of an embodiment of an egress unit which is an example of a frame output device. In FIG. 5, the frame is input from the switching unit 2 and supplied to the header processing unit 21. The header processing unit 21 adds the in-device header shown in FIG. 6 to the head of the frame.

  In FIG. 6, the in-device header is composed of, for example, 16 bytes, the output class and the in-device class are in the first byte, and the frame length as length information is in the 8th to 9th bytes. The 15th to 16th bytes include a destination physical port and a destination logical port as destination queue information.

  The frame with the in-device header output from the header processing unit 21 is supplied to the buffer control unit 22, and the destination queue information and the frame length extracted from the in-device header are supplied to the pointer management unit 23.

  The pointer management unit 23 generates a write pointer as a frame write address and a read pointer as a read address for the frame buffer 25. The write pointer issuing unit 23b of the pointer management unit 23 accesses the tail pointer management memory 23a and the chain management memory 24, and issues a write pointer in units of pages of a predetermined length (for example, 256 bytes) to the queue indicated by the destination queue information. To the write control unit 22 a of the buffer control unit 22.

  The tail pointer management memory 23a manages the last pointer of the queue for each queue, and the chain management memory 24 manages the queue pointer chain and frame length for each queue. As a result, the write control unit 22a writes a frame in page units in the frame buffer (memory) 25 using the write pointer as an address.

  The write pointer issuing unit 23b supplies the enqueue information, which is frame input information, to the frame accumulation counter 26a of the scheduler unit 26 by issuing the write pointer. Thereby, the frame accumulation counter 26a increments the frame accumulation number of the queue among the frame accumulation numbers counted for each queue and notifies the round robin unit 26b.

  The scheduler unit 26 adjusts the reading order of frames from the frame buffer 25. The token counter 26c of the scheduler unit 26 counts tokens for each queue. A fixed amount of tokens is periodically added, and the frame length of the frame is subtracted every time a frame is read from the queue.

  The round robin unit 26b is notified of a token for each queue, that is, token presence / absence information from the token counter 26c, and is notified of the number of frames stored for each queue from the frame storage counter 26a. Based on these, the round robin unit 26b determines a queue to be read by the round robin method. The round robin unit 26b uses a queue having both a token and a frame accumulation number (a token count value is a positive value and a frame accumulation number is a positive value) as a round robin entry candidate, and the queue priority is selected from the entry candidates. The queue to be read is determined according to the degree.

  At this time, the token presence / absence information of the token counter 26c used by the round robin unit 26b has not yet been read out for the previously selected read request, and is still stored in the read request FIFO 23e of the pointer management unit 23. Therefore, it is the value before confirmation. Then, the round robin unit 26b supplies the read request to the pointer management unit 23 and the determined queue information to the read request FIFO 23e, and decrements the frame accumulation value of the frame accumulation counter 26a for the queue corresponding to the read request.

  The queue information output from the round robin unit 26b is stored in the read request FIFO 23e of the pointer management unit 23. The read request FIFO 23e is configured to store n (n is a fixed value of, for example, about 2 to 10) read requests. When the read request FIFO 23e is about to overflow, the read request FIFO 23e is backed up to the round robin unit 26b. A pressure is issued, whereby the round robin unit 26b stops the round robin operation.

  The read pointer issuing unit 23c of the pointer management unit 23 periodically retrieves a read request from the read request FIFO 23e, accesses the head pointer management memory 23d and the chain management memory 24 for the queue that is the target of the retrieved read request, and obtains the read pointer. Issue. The head pointer management memory 23d manages the head pointer of the queue that is the target of the read request, and the read pointer issuing unit 23c traces the pointer chain of the chain management memory 24 from the head pointer of the head pointer management memory 23d to page unit. Issue a read pointer.

  The read pointer issuing unit 23c supplies the issued read pointer as an address to the read control unit 22b of the buffer control unit 22. The read control unit 22b reads a frame from the frame buffer 25 in units of pages.

  The read pointer issuing unit 23c subtracts the frame length stored in the chain management memory 24 from the token counter 26c when issuing the read pointer of the head pointer of the queue that is the target of the read request. In this embodiment, since the operation of the scheduler unit 26 precedes the operation of the pointer management unit 23 and outputs a read request, subtraction is permitted even if the count value of the token counter 26c becomes a negative value.

  The frame read from the frame buffer 25 is temporarily stored in a buffer for each port of the output buffer unit 27, and output from each buffer to the outside.

  As described above, the read request FIFO 23e for accumulating the read requests from the scheduler unit 26 is provided in the pointer management unit 23, the processing of the scheduler unit 26 is performed prior to the processing of the pointer management unit 23, and the read request is sent to the read request FIFO 23e. Accumulate.

  Further, when the read request accumulation number of the read request FIFO 23e exceeds a predetermined value and the read request FIFO 23e is about to overflow, a back pressure is output to the scheduler unit 26 to stop scheduling.

  As a result, the scheduler unit 26 can be operated in advance before the token presence / absence information is determined, and the next scheduling process is performed in advance without waiting for the completion of frame reading, thereby outputting a read request at a constant cycle. It becomes possible to do.

  Also, the scheduler unit 26 is provided with a frame accumulation counter 26a, and decrements the frame accumulation counter 26a simultaneously with the read request output. This makes it possible to update the frame presence / absence information, which is the count value of the frame accumulation counter 26a, before accessing the chain management memory 24 and confirming the next frame (next chain), and pointer management for managing the pointer chain The unit 23 and the scheduler unit 26 that selects the queue to be read can be operated independently without affecting each other. Also, the fact that the frame presence / absence information can be determined early means that the timing for operating the scheduler unit 26 can be made earlier. Accordingly, there is a margin in the scheduling process that must be performed in a certain time.

<Signal timing chart>
FIG. 7 shows a signal timing chart of each part of the circuit of FIG. Here, the frame readout cycle T3 is fixed. Assuming that the number of accumulated frames = 3 and the token = + 50 [bytes] for the specific queue, a frame read request is output at time t11, for example, an output of a frame of frame length = 100 [bytes] is instructed. As a result, the number of accumulated frames = 2 at time t12. However, since the frame is not actually read from the specific queue at this time, the token remains at +50.

  At time t13, a frame read request is output, and for example, output of a frame having a frame length of 100 [bytes] is instructed. As a result, the number of accumulated frames = 1 at time t14. At this point in time, since the frame is not actually read from the specific queue, the token remains at +50.

  At time t15, a frame read request is output, and for example, output of a frame with a frame length = 100 [bytes] is instructed. As a result, the number of accumulated frames = 0 at time t16. At this point in time, since the frame is not actually read from the specific queue, the token remains at +50.

  At time t17, the frame of the frame read request (instruction at time t11) is actually output, and token = −50. At time t18, since the number of accumulated frames = 0 at the previous time t16, no frame read request is output.

  Thereafter, the frame of the frame read request (instruction at time t13) is actually output at time t19, and token = −150. At time t20, the frame of the frame read request (instruction at time t15) is actually output, and token = −250.

Conventionally, unless frame information is read from the chain management memory 24, the token presence / absence information and frame presence / absence information necessary for scheduling cannot be updated. On the other hand, in the present embodiment, the pointer management unit 23 and the scheduler unit 26 are configured to operate independently without affecting each other. That is, the scheduler unit 26 manages the token presence / absence information and the frame presence / absence information independently by using the frame accumulation counter 26a and the token counter 26c. As a result, there is a margin in the scheduling process, the logic load can be reduced in configuring the circuit, and it is possible to cope with an increase in the number of queues in the future.
(Appendix 1)
Pointer management means for generating an address for writing the input frame to the frame buffer for each queue corresponding to the destination, and generating an address for reading the frame from the frame buffer;
Scheduler means for adjusting a frame reading order from the frame buffer and generating a frame reading request for each queue;
Request buffer means for accumulating frame read requests generated by the scheduler means, sequentially taking out the stored frame read requests and supplying them to the pointer management means;
Have
A frame output device for outputting a frame read from the frame buffer at an address generated by the pointer management means.
(Appendix 2)
In the frame output device according to attachment 1,
The frame output apparatus characterized in that the request buffer means fetches stored frame read requests periodically and in the order of storage.
(Appendix 3)
In the frame output device according to attachment 2,
A frame output device comprising: an accumulation number counting means for counting the number of accumulated frames for each queue of the frame buffer and transmitting the counted number to the scheduler means.
(Appendix 4)
In the frame output device according to attachment 3,
The accumulation number counting means increments the frame accumulation number for each queue when generating an address for writing the frame input by the pointer management means to the frame buffer for each queue, and queues it according to the frame read request generated by the scheduler means. A frame output device that decrements the number of accumulated frames every time.
(Appendix 5)
A first step of generating an address for writing an input frame to a frame buffer for each queue corresponding to a destination;
A second step of adjusting a frame reading order from the frame buffer to generate a frame reading request for each queue;
A third step of accumulating the frame read requests generated in the second step and sequentially taking out the accumulated frame read requests;
A fourth step of generating an address for reading a frame from the frame buffer in response to the frame read request extracted in the third step;
Have
A frame output method comprising: outputting a frame read from the frame buffer at the address generated in the fourth step.
(Appendix 6)
In the frame output method according to attachment 5,
3. The frame output method according to claim 3, wherein in the third step, stored frame read requests are taken out periodically and in the order of storage.
(Appendix 7)
In the frame output method according to attachment 6,
A frame output method comprising a fifth step of counting the number of accumulated frames for each queue of the frame buffer and transmitting it to the second step.
(Appendix 8)
In the frame output method according to appendix 7,
In the fifth step, when an address for writing the frame input in the first step to the frame buffer is generated for each queue, the frame accumulation number is incremented for each queue, and the frame is read by the frame read request generated in the second step. A frame output method comprising decrementing the number of accumulated frames every time.
(Appendix 9)
In the frame output device according to attachment 4,
The request buffer means stops the operation of the scheduler means when the accumulated number of frame read requests exceeds a predetermined value.

DESCRIPTION OF SYMBOLS 1 Ingress part 2 Switching part 23 Egress part 21 Header processing part 22 Buffer control part 22a Write control part 22b Read control part 23 Pointer management part 23a Tail pointer management memory 23b Write pointer issue part 23c Read pointer issue part 23d Head pointer management memory 23e Read request FIFO
24 Chain management memory 25 Frame buffer 26 Scheduler unit 26a Frame accumulation counter 26a
26b Round robin 26c Token counter 27 Output buffer

Claims (4)

Pointer management means for generating an address for writing the input frame to the frame buffer for each queue corresponding to the destination, and generating an address for reading the frame from the frame buffer;
Scheduler means for adjusting a frame reading order from the frame buffer and generating a frame reading request for each queue;
Request buffer means for accumulating frame read requests generated by the scheduler means, sequentially taking out the stored frame read requests and supplying them to the pointer management means;
An accumulated number counting means for counting the number of accumulated frames for each queue of the frame buffer and transmitting it to the scheduler means;
Have
A frame output device for outputting a frame read from the frame buffer at an address generated by the pointer management means. The frame output device according to claim 1, wherein
The frame output apparatus characterized in that the request buffer means fetches stored frame read requests periodically and in the order of storage. The frame output device according to claim 2 , wherein
The accumulation number counting means increments the frame accumulation number for each queue when generating an address for writing the frame input by the pointer management means to the frame buffer for each queue, and queues it according to the frame read request generated by the scheduler means. A frame output device that decrements the number of accumulated frames every time. A first step of generating an address for writing an input frame to a frame buffer for each queue corresponding to a destination;
A second step of adjusting a frame reading order from the frame buffer to generate a frame reading request for each queue;
A third step of accumulating the frame read requests generated in the second step and sequentially taking out the accumulated frame read requests;
A fourth step of counting the number of accumulated frames per queue of the frame buffer and communicating it to the second step;
A fifth step of generating an address for reading a frame from the frame buffer in response to the frame read request fetched in the third step;
Have
A frame output method comprising: outputting a frame read from the frame buffer at the address generated in the fifth step.

Images