KR20170100414A - Memory controller and request scheduling method using the same - Google Patents

Abstract

The present invention relates to a memory controller which comprises: a plurality of request queues for storing a request transmitted from a corresponding host device among a plurality of host devices; a token information generating unit for generating information on the number of first and second tokens corresponding to each of the plurality of host devices; and a request scheduler for sequentially selecting by repeating the plurality of request queues, outputting requests stored in the selected request queue by using the first and second tokens, outputting one or more requests per one first token, and outputting one per one second token when the first tokens are all consumed. The memory controller can optimize an output order of the plurality of host devices by determining an output order of the stored requests by using two different types of tokens assigned to each host device.

Description

[0001] MEMORY CONTROLLER AND REQUEST SCHEDULING METHOD USING THE SAME [0002]

This patent document relates to a request scheduling method of a memory controller and a memory controller.

The memory controller can determine the processing order of a plurality of requests provided from the host device and output the determined order to the memory device. Conventional memory controllers use a scheduling method according to rules such as FCFS (First Come First Served) and FR-FCFS (First Ready Frist Come First Served) in order to determine a processing order of a plurality of requests. Thus, the conventional memory controller can perform scheduling based on a fixed reference, but can not provide optimized scheduling according to the characteristics of the host apparatus.

The present invention provides a memory controller that can optimize the output order of requests of a plurality of host devices by determining the output order of stored requests using two different types of tokens allocated to each host device, A scheduling method can be provided.

A memory controller according to an embodiment of the present invention includes a plurality of request queues for storing requests transmitted from corresponding host devices among a plurality of host devices; A token information generator for generating information on the number of first and second tokens corresponding to the plurality of host devices, respectively; And outputting one request per one of the first tokens, and outputting a request for one of the plurality of request queues, And a request scheduler outputting one request per one of the second tokens if the tokens are exhausted.

A method of requesting a memory controller to output a request transmitted from a plurality of host devices to a memory device, the method comprising the steps of: Assigning a second token; Storing a request transmitted from the plurality of host devices in a request queue corresponding to each of the plurality of host devices; Selecting a request queue of one of the plurality of request queues; Outputting the requests stored in the selected request queue for each of the first tokens one by one; And outputting, if the first token is consumed, the requests stored in the selected request queue per one of the second tokens, one by one.

This technique allocates two different types of tokens according to the characteristics of each host device and determines the output order of requests transmitted from each host device using the tokens, thereby enabling a maximized request scheduling suitable for each host device .

1 is a block diagram of a system 100,
FIG. 2 is a view for explaining an example of a method of scheduling a request considering two or more of the above-mentioned rules; FIG.
FIG. 3 is a diagram for explaining a problem of the scheduling method illustrated in FIG. 2,
Figure 4 illustrates a memory controller 400 and peripheral configuration in a system according to one embodiment of the present invention,
5 is a diagram for explaining the operation of the memory controller 400 of FIG. 4,
Figure 6 illustrates a memory controller 600 and peripheral configuration in a system according to one embodiment of the present invention,
FIG. 7 is a diagram for explaining the operation of the memory controller 600 of FIG. 6,
8 illustrates a memory controller 700 and peripheral configuration in a system according to an embodiment of the present invention,
FIG. 9 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention. FIG.
FIG. 10 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention. FIG.
11 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention.
12 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to facilitate a person skilled in the art to easily carry out the technical idea of the present invention.

1 is a block diagram of the system 100. As shown in FIG.

Referring to FIG. 1, the system may include a plurality of host devices 110_0 through 110_n, where n is a natural number, a memory controller 120, and a memory device 130.

The plurality of host devices 110_0-110_n are devices such as a central processing unit (CPU), a graphic processing unit (GPU) or a display device and may make requests to the memory controller 120 .

The memory controller 120 may schedule requests sent from the host devices 110_0-110_n to control the memory device 130 and transfer data between the memory device 130 and the plurality of host devices 110_0-110_n have. The memory controller 120 may include a bus interface 121, a plurality of request queues 122_0 - 122_n, and a request scheduler 123.

The bus interface 121 communicates the requests sent from the host devices 110_0-110_n to the corresponding request queues 122_0-122_n and between the host devices 110_0-110_n and the memory device 130 As shown in FIG.

Each of the plurality of request queues 122_0 to 122_n corresponds to a host device of one of the plurality of host devices 110_0 to 110_n and transmits requests transmitted from the corresponding host device through the bus interface 121 in the inputted order Can be stored.

The request scheduler 123 may schedule requests stored in the plurality of request queues 122_0 through 122_n according to a set rule and output the requests to the memory device 130 in a scheduled order.

The memory device 130 may store the data transferred through the bus interface 121 or output the stored data according to a request transmitted from the memory controller 120. [ The memory device 130 may include a plurality of memory banks (BK0 - BKm, where m is a natural number) for storing data. Each of the plurality of memory banks BK0 to BKm may include a plurality of word lines WL, bit lines BL and memory cells MC. In FIG. 1, only a part of the structure is shown inside the memory bank BK0 for convenience of illustration.

Since the quality of the QoS (Quality of Service) required by the host device is different depending on the type of the host device, the memory controller 120 considers the characteristics of the host devices, ). ≪ / RTI > For example, a central processing unit may require a low latency, a display unit may require a high bandwidth, a medium latency, and a graphics processing unit may require a high bandwidth. In this case, if the requests transmitted from the different host devices are transmitted to the memory device 130 in the requested order, the operation corresponding to the request may not be efficiently performed.

As described above, there are (1) a method using a timeout rule, (2) a method using a priority rule, or (3) a method using a bandwidth calibration rule.

(1) The method of using the timeout rule may be a method of assigning different threshold time to each host device and outputting requests that are received among the requests stored in the request queue or exceeding the threshold time. For example, when a threshold time of 5 clocks is allocated to the host apparatus A and a threshold time of 10 clocks is allocated to the host apparatus B, the request sent from the host apparatus A is output when the memory controller receives 5 clocks or more, The request transmitted from the host apparatus B can be output when it is received by the memory controller for 10 clocks or more.

(2) The method using the priority rule is a method of outputting a request having a high priority in consideration of a priority given to a request. The priority given to the request may be fixedly set for each host device, or may be differently assigned to each request. For example, when a higher priority is given to the host apparatus A than to the host apparatus B, a request from the host apparatus A and the host apparatus B can be simultaneously output.

(3) The method of using the bandwidth calibration rule sets the bandwidth to be allocated for each host device (hereinafter referred to as the target bandwidth), monitors the bandwidth actually used for each host device, and as a result, ) Is greater than the target bandwidth, the priority for the host device may be lowered, and if the actual bandwidth is smaller than the target bandwidth, the priority for the host device may be increased. For example, if the bandwidth used by the host apparatus A and the bandwidth used by the host apparatus B are monitored during the set interval and compared with the target bandwidth of each host apparatus, if the bandwidth of the host apparatus A is larger than the target bandwidth, The priority of the host apparatus B can be lowered and the priority of the host apparatus B can be increased if the bandwidth of the host apparatus B is smaller than the target bandwidth.

(4) The method using the FCFS (first come-first serve) rule may be a method of outputting the received request first.

(5) The method of using the first-ready (FR-FCFS) rule is to first output the received request, if there is a currently open (active) row (word line) It may be a method of outputting the request first.

(6) The method of using a first-ready (FR) rule may be a method of first outputting a request for such a row if there is a currently open (or active) row (or word line).

At this time, when scheduling actual requests, scheduling may be performed considering only one of the above rules, but scheduling may be performed by considering two or more rules in combination.

FIG. 2 is a view for explaining an example of a method of scheduling a request considering two or more of the above-mentioned rules.

Referring to FIG. 2, the output order can be determined according to four items 'TIMEOUT', 'PRIORITY', 'ROWHIT', and 'STARVATION' in each of the requests CMD0 to CMD3.

'TIMEOUT' indicates whether the elapsed time from the receipt of each request is greater than or equal to the threshold time allocated to each request, 'T' indicates that the elapsed time is greater than or equal to the threshold time, 'F' May indicate that the elapsed time is less than the threshold time. At this time, the request whose 'TIMEOUT' is 'T' can be outputted earlier than the request whose 'T' is 'F'.

'PRIORITY' indicates the priority score assigned to each request, and a higher priority request can be given a higher priority score. At this time, requests with higher priority scores can be output faster than lower ones.

'ROWHIT' indicates whether the row corresponding to the request is currently open, 'T' indicates that the row corresponding to the request is currently open, and 'F' indicates that the row corresponding to the request is currently open . At this time, the request whose 'ROWHIT' is 'T' can be output earlier than the request whose 'R' is 'F'.

'STARVATION' indicates the delay time in which the received request can not be output, and the long delay request can be output faster than the short delay request.

In FIG. 2, weights are given in order of 'TIMEOUT', 'PRIORITY', 'ROWHIT', and 'STARVATION'. Therefore, 'TIMEOUT' is considered the most important in scheduling, 'PRIORITY' is considered next, 'ROWHIT' is considered next, and 'STARVATION' is considered last.

For example, if the 'TIMEOUT' of the two requests is different, the 'T' request with 'TIMEOUT' may be output before the request with 'F', regardless of the other three. Next, if the 'TIMEOUT' of the two requests are the same (all 'T' or all 'F') and the 'PRIORITY' is different, the request with higher PRIORITY may be output before the lower request, have. Next, if 'TIMEOUT' and 'PRIORITY' of two requests are the same, and 'ROWHIT' is different, 'ROWHIT' may be output before the 'T' request is 'F', regardless of the remaining one item. Finally, if 'TIMEOUT', 'PRIORITY' and 'ROWHIT' are the same, and 'STARVATION' is different, a request with a higher 'STARVATION' may be output before a lower request.

Taking this into consideration, when the requests CMD0 to CMD3 shown in FIG. 2 are scheduled, the order of CMD0 to CMD1 to CMD2 to CMD3 can be obtained. At this time, the item to be considered as a priority among the plurality of items (that is, the rule to be considered as priority) may be different depending on the design.

FIG. 3 is a diagram for explaining a problem of the scheduling method illustrated in FIG.

Referring to FIG. 3, the request scheduler 123 compares 'TIMEOUT', 'PRIORITY', 'ROWHIT', and 'STARVATION' of the requests CMD0 to CMD3 stored in the request queues 122_0 to 122_n, (OUT_CMD) to be finally output.

As shown in FIG. 3, the scheduling method considering all the items 'TIMEOUT', 'PRIORITY', 'ROWHIT' and 'STARVATION' has an advantage that all requests (CMD0 to CMD3) However, as the number of requests (CMD0 - CMD3) increases, the number of comparisons increases, which slows down the operation and increases the circuit area required for such comparison.

4 is a diagram illustrating a memory controller 400 and a peripheral configuration in a system according to an embodiment of the present invention.

4, the system may include a plurality of host devices (HOST_0 to HOST_n), a memory controller 400, and a memory device (MEMORY). The memory controller 400 may include a plurality of request queues 410_0-410_n, a token information generator 420, a request scheduler 430 and a bus interface 440. [

A plurality of host devices HOST_0 to HOST_n are connected to a central processing unit (CPU), a graphic processing unit (GPU), a display device, a video device VIDEO, an audio device AUDAMA, ), And may transmit requests and data to the memory controller 400 or receive data transmitted from the memory controller 400 as needed.

The memory device MEMORY may store the data transferred through the bus interface 440 or output the stored data according to the request transmitted from the memory controller 400. [ The memory device MEMORY may include a plurality of memory banks (BK0 - BKm, where m is a natural number) for storing data. Each of the plurality of memory banks BK0 to BKm may include a plurality of word lines (WL, row), bit lines (BL, column) and memory cells MC (see BK0 in Fig. 1).

The memory controller 400 may receive a read or write request from the host devices HOST_0 to HOST_n and control the memory device MEMORY to perform the requested operation in response thereto. To this end, the memory controller 400 can schedule requests transmitted from a plurality of host devices (HOST_0 to HOST_n) and output them to the memory device (MEMORY). The memory controller 400 may also transfer the data transmitted from the host devices HOST_0 to HOST_n to the memory device MEMORY or may transfer the data transmitted from the memory device MEMORY to the host devices HOST_0 to HOST_n .

The plurality of request queues 410_0 through 410_n may store requests transmitted from the corresponding one of the plurality of host devices HOST_0 through HOST_n in the order in which they are received. The multiple request queues 410_0-410_n may output the stored requests to the request scheduler 430 (CMDs_0 - CMDs_n).

The token information generating unit 420 may generate and output information (TK0-TKn, hereinafter referred to as token information) about the first and second number of tokens corresponding to the host apparatuses HOST_0 to HOST_n, respectively. In the following, the token may be the price used to output the request. For example, one of the one or more tokens assigned to the host device A may need to be consumed to output the request of the host device A. [ The first token and the second token may be used according to another of the above-mentioned rules. When outputting a request using the first token, the request may be output according to the FR-FCFS rules. The second token is a token used after all of the first token is consumed, and the request may be output according to the FR rule when outputting the request using the second token.

When the first or second token is used and the request is output, the token information generator 420 reflects the token information (TK0 - TKn), and when the order of outputting the request of the host device is ended, Can be initialized. Therefore, when the next request queue of the selected request queue is selected, the token information generation unit 420 can initialize the token information (TK0 - TKn) corresponding to the previously selected request queue. That is, if a host device is selected and the requests are output using the first or second token and the next host device is selected, the number of first and second tokens of the previously selected host device may be returned to the initial state .

The number of first and second tokens allocated to the plurality of host devices HOST_0 to HOST_n may be determined according to characteristics required for the system or characteristics of each host device. For example, when a high performance is required for a system, an intermediate number of first tokens and a large number of second tokens are assigned to each host device, and when high fairness is required, A small number of first tokens and an intermediate number of second tokens may be assigned to the second token. In addition, when a host device such as a CPU requiring a high QoS (Quality of Service) and a host device such as a GPU requiring a high bandwidth coexist in the system, a host device such as a CPU has a large number of first tokens and a small number of 2 token, and a host device such as a GPU may be assigned a small number of first tokens and an intermediate number of second tokens.

The request scheduler 430 repeatedly selects a plurality of request queues 410_0-410_n one by one, and outputs the requests stored in the selected request queue using the first and second tokens, one request per one token And output one request per one of the second tokens if the first token is exhausted. At this time, the request scheduler 430 may preferentially use the first token to output the request, and output the request using the second token only after all of the first token is used.

The request scheduler 430 determines whether the first or second token remains, but there is no request to output to the selected request queue any more, or the first token If there is no request to do so, the next request queue of the selected request queue can be selected.

The request scheduler 430 sequentially outputs requests in accordance with a first-ready (first-come-first-serve) (FR-FCFS) rule when outputting a request using a first token and outputs a request using a second token , The request can be output in order according to the first-ready (FR) rule. When determining the order of the request according to the FR-FCFS rule or the FR rule, it may be considered whether or not it is the same as the row address of the memory device corresponding to the immediately preceding request. That is, when a request is selected by the FR rule, considering whether there is a request corresponding to the row currently opened, if there is more than one request corresponding to the opened row, , And may not print requests if they do not exist.

The request scheduler 430 may skip the selection of the request queue among the plurality of request queues in which the request is not stored and select the next request queue. The request scheduler 430 may select a request queue according to a round robin scheme.

The bus interface 440 communicates the request sent from the host devices HOST_0 through HOST_n to the corresponding request queues 410_0 through 410_n and sends the data sent from the plurality of host devices HOST_0 through HOST_ (MEMORY) or transfer data from the memory device (MEMORY) to a plurality of host devices (HOST_0 to HOST_n).

The number of first and second tokens assigned to the plurality of host devices HOST_0 to HOST_n may be fixed, but may change as needed during operation of the system. The system can easily trade-off between high performance and high fairness by scheduling requests using two different tokens. Therefore, the token can be set for each host device according to the needs of the system, so that QoS operation can be performed accurately and flexibly.

5 is a diagram for explaining the operation of the memory controller 400 of FIG.

5, the system includes two host devices (HOST_0, HOST_1), two first and second tokens are assigned to the first host device HOST_0, respectively, and a second host device The first and second tokens are assigned to one and four HOST_1, respectively.

'HO' denotes a request queue corresponding to the first host device HOST_0, and contents described in each column of the request queue are transmitted from the first host device HOST_0 to the stored request, It may be the request sent.

'H1' denotes a request queue corresponding to the second host device HOST_1, and contents described in each column of the request queue are transmitted from the second host device HOST_1 to the stored request, It may be the request sent.

'TK1_0' may be the number of the first tokens corresponding to the first host device HOST_0 and 'TK2_0' may be the number of the second tokens corresponding to the first host device HOST_0. Also, 'TK1_1' may be the number of the first tokens corresponding to the second host device HOST_1, and 'TK2_1' may be the number of the second tokens corresponding to the second host device HOST_1.

For reference, the request consists of two characters, the first of the two letters represents the row address corresponding to the request, and the following letter can represent the column address corresponding to the request. For example, a request A1 may be a request for memory cells corresponding to a row address A and a column address 1.

Hereinafter, the scheduling operation of the request scheduler 430 may be described on the assumption that the request queues corresponding to the first and second host devices HOST_0 and HOST_1 are alternately selected. It can also be assumed that requests are not sent from the host devices HOST_0 and HOST_1 until all requests stored in the request queues HO and H1 are outputted in the current state.

First, the request queue HO is selected because there is a request in the request queue HO, one first token of the first host device HOST_0 is used, and the request A1 is output. Next, the first token of the first host device (HOST_0) is used, and the first received request (A2) among the requests (A2 - A4) having the same row address as the request (A1) may be output. In order to output the request received from the first host device HOST_0 using the second token since the first token of the first host device HOST_0 is exhausted, There may be a request corresponding to a row. A request corresponding to an opened row may indicate a request that has the same row address as a recently issued request.

Since the row address of the recently output request A2 is A and the requests A3 and A4 whose row address is A among the requests stored in the request queue HO are present, , And then A4 can be output using the second token. After the request A4 is outputted, there is no request with the corresponding row address A in the request queue HO, and since the first token corresponding to the first host device HOST_0 is all consumed, Requests sent from one host device (HOST_0) can no longer be output. Thus, the request queue HO is deselected, and the next request queue H1 can be selected. When the next request queue H1 is selected, the token information of the previously selected request queue H0 can be initialized. That is, the number of the first and second tokens corresponding to the first host device HOST_0 may be initialized to 2 and 4, respectively.

The request stored in the request queue H1 may then be output using the first and second tokens allocated to the second host device HOST_1. First, a request (C1) may be output using the first token. If the first token is exhausted, the request sent from the second host device HOST_1 may be output using the second token. In the case of using the second token, only the request which is the same as the recently outputted request (C1) and the corresponding row address can be output. Therefore, the request C2 is output using the second token, and the request queue H0 can be selected again because there is no request corresponding to the row that is no longer open. At the same time, the number of the first and second tokens corresponding to the second host apparatus HOST_1 can be initialized.

Next, the requests transmitted from the first host device HOST_0 may be output. The first token-request B1, the first token-request B2, the second token-request B3, the second token-request B4, and the token- Can be output in order. Since there is no more stored request in the request queue H0, the request queue H1 may be selected and the requests sent from the second host device HOST_1 may be output. At this time, the number of the first and second tokens corresponding to the first host device HOST_0 may be initialized.

Next, the token used for the output of the requests transmitted from the second host device H1 and the output request are listed in order. The first token-request D1, the second token-request D2, 2 token-request D3, second token-request D4, second token-request D5. Since both the first and second tokens are exhausted, the next request queue H0 must be selected. However, since there is no more stored request in the request queue H0, the selection of the request queue H0 is skipped and the request queue H1 can be selected again. At this time, since the request queue H1 is reselected, when the selection of the request queue H1 is canceled in the previous selection, the number of the first and second tokens corresponding to the second host device HOST_1 may be initialized . Thus, the request D6 may be output using the first token.

As described above, the memory controller 600 repeatedly selects a plurality of request queues (HO, H1), sequentially selects requests stored in the selected request queue according to the FR-FCFS rule using the first token, If you have used all the tokens, you can use the second token to select and output according to the FR rules.

6 is a diagram illustrating a memory controller 600 and peripheral configuration in a system according to an embodiment of the present invention.

Referring to FIG. 6, the system may include a plurality of host devices HOST_0 to HOST_n, a memory controller 600, and a memory device (MEMORY). The memory controller 600 may include a plurality of request queues 610_0-610_n, a token information generator 620, a request scheduler 630 and a bus interface 640. [

In the system of FIG. 6, there may be a priority-host device among a plurality of host devices HOST_0 to HOST_n unlike the system of FIG. Priority - A host device is a host device that requires a fast response to a request, for example, a host device that requires a low latency, such as a CPU.

The configuration and operation of the memory controller 600 of FIG. 6 may basically be similar to the configuration and operation of the memory controller 400 of FIG.

However, the request scheduler 630 of the memory controller 600 of FIG. 6 is configured such that when one or more of the host devices HOST_0 to HOST_n are set as a priority-host device, Interrupt the operation of outputting requests of the request queue, select the request queue corresponding to the priority-first host device, and output the requests.

To this end, a request queue corresponding to a priority-host device among a plurality of request queues can activate a TOP_PRIORITY signal when a request is received. The request scheduler 630 may pause the request output operation that was in progress when the highest priority signal TOP_PRIORITY is active and may resume outputting the request that was paused after first outputting the request received in the request queue corresponding to the highest priority host device have. The priority-first host device can be fixedly or dynamically set as one or more host devices of the plurality of host devices (HOST_0 - HOST_n).

7 is a diagram for explaining the operation of the memory controller 600 of FIG.

7, the system includes three host devices (HOST_0 to HOST_2), two first and second tokens are assigned to the first host device HOST_0, respectively, and a second host device One and two tokens are allocated to the third host device HOST_2, one and two tokens are assigned to the third host device HOST_2, and one and two tokens are allocated to the third host device HOST_2, A scheduling operation in the case of a priority-host device can be described.

'TK1_2' is the number of first tokens corresponding to the third host device HOST_2, and 'TK2_2' is the number of second tokens corresponding to the third host device HOST_2.

Assume that the requests stored in the request queues HO - H2 in the initial state are as shown in FIG. At this time, it is assumed that the request is stored in the request queue H2. In this case, since the selection for the request queue H2 is skipped, the order in which the requests are output may be the same as shown in FIG. Assume that a request is sent from a host device (HOST_2), which is a priority-host device, while outputting requests. The request scheduler 630 may pause the operation of scheduling and outputting the requests of the currently selected request queue and output the requests E1 and E2 sent from the host device HOST_2. The output operation of the interrupted request queue can be resumed (RESUME) when the output of the requests E1 and E2 transmitted from the host device HOST_2 is completed.

8 is a diagram illustrating a memory controller 700 and a peripheral configuration in a system according to an embodiment of the present invention.

8, the system may include a plurality of host devices (HOST_0 to HOST_n), a memory controller 800, and a memory device (MEMORY). The memory controller 800 includes a plurality of request queues 810_0 to 810_n, a token information generating unit 820, a request scheduler 830, a bus interface 840, a bandwidth monitoring unit 850 and a bandwidth adjusting unit 860 .

In the system of FIG. 8, the memory controller 800 may include a bandwidth monitoring unit 850, unlike the system of FIG. The configuration and operation of the memory controller 800 of FIG. 8 may be basically similar to the configuration and operation of the memory controller 400 of FIG.

However, the bandwidth monitoring unit 850 of the memory controller 800 of FIG. 8 can monitor the bandwidth between the plurality of host devices HOST_0 to HOST_n and the memory controller 800. In more detail, the bandwidth monitoring unit 850 detects the number of requests output through the request scheduler 830 among the requests stored in the request queues 810_0 - 810_n during the set intervals, The actual bandwidth for the host device (HOST_0 - HOST_n) can be measured.

The bandwidth control unit 860 refers to the monitoring result of the bandwidth monitoring unit 850 and determines the number of first or second tokens allocated to the host apparatus whose actual bandwidth is smaller than the target bandwidth among the plurality of host apparatuses HOST_0 to HOST_n And increase the number of first or second tokens assigned to host devices whose actual bandwidth is greater than the target bandwidth. The bandwidth control unit 860 controls the bandwidth control unit 860 to reduce the number of the corresponding first or second tokens in the case of a request queue having a smaller number of requests output through the request scheduler 830 during a predetermined interval than the target value 820 and controls the token information generator 820 to increase the number of corresponding first or second tokens in the case of a request queue in which the number of requests outputted through the request scheduler 830 is smaller than the target value for a predetermined interval, Can be controlled.

The memory controller can maximize the efficiency of the system by adjusting the number of tokens allocated to each host device based on the comparison of the actual bandwidth and the target bandwidth.

The scheduling method of the memory controller can be divided into two steps. First, the first step is to select a request queue, i.e., a host device, of a plurality of request queues. The second step is to output the request using the first and second tokens with the request queue selected. Hereinafter, the first step and the second step described above can be separately described in FIG. 9 and FIG. 10, respectively.

9 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention. 9 is a diagram for explaining a first step of the scheduling method of the memory controller and a first step of the second step, that is, a step of selecting a request queue (or a host device).

9, the scheduling method of the memory controller includes a token allocation step S910, a host device designation step S920, a host device selection step S930, a request output step S940, and a request reception step S950 can do.

The token allocating step S910 is performed before selecting the host device and outputting the request. The number of the first and second tokens in the token allocating step S910 is determined according to the characteristics of the host device or the requirements of the system Can be determined accordingly. The details of the token allocation can be the same as described above in the description of FIG.

The host device designation step (S920) may be a step of designating a next host device to be selected among a plurality of host devices. In the host device designation step S920, a host device to be selected next is specified (S921), and it is determined whether a request is stored in a request queue corresponding to the host device to be selected next (S922) If the request is stored in the corresponding request queue (YES), the process proceeds to the host device selection step (S930) to select the request queue corresponding to the specified host device, and the request is stored in the request queue corresponding to the host device to be selected next (NO), the flow advances to step S921 to designate another host device.

In the host device selection step (S930), a request queue corresponding to the designated host device can be selected.

In the request output step S940, the requests stored in the selected request queue may be outputted according to the order given by the request scheduler. In the request output step S940, each request may be output using the first or second token.

For reference, the request output step S940 may include a part or all of the section in which the host apparatus designation step S920 is performed simultaneously or the request outputting step S940 is performed, ≪ / RTI >

It is also possible to send a request transmitted from a plurality of host devices to a plurality of host devices independently of or in association with the token allocation step S910, the host device designation step S920, the host device selection step S930 and the request output step S940 A request reception step (S950) may be performed in which each is stored in a corresponding request queue. The request reception step S950 may be performed in a different interval from the token allocation step S910, the host device designation step S920, the host device selection step S930, or the request output step S940. A part or all of the section in which the request reception step S950 is performed may be performed during the period in which the token allocation step S910, the host device designation step S920, the host device selection step S930, or the request output step S940 is performed Some or all of which may overlap.

10 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention. 10 is a diagram illustrating a step of outputting a request using the first and second tokens in a first step of the scheduling method of the memory controller and a second step of the second step, that is, a request queue is selected.

10, a method of scheduling a memory controller includes a first token checking step S1010, a first open row detecting step S1020, a first request outputting step S1030, a first token reducing step S1040, 2 token verification step S1050, a second open row detection step S1060, a second request output step S1070, a second token reduction step S1080, a selection end and a token initialization step S1090 .

In the first token check step (S1010), it is possible to check whether there is a first token available for outputting a request of the host apparatus corresponding to the selected request queue. If the number of remaining first tokens in the first first token check step S1010 is greater than 0, the process proceeds to the first open row detecting step S1020 (YES). If the number of remaining first tokens is 0 And proceeds to the second token checking step S1050 (NO).

In the first open row detection step (S1020), if there is a request corresponding to the open row among the requests stored in the request queue, the process proceeds to 'YES' (S1031). If there is no request corresponding to the open row among the requests stored in the request queue, the process proceeds to NO to determine whether the request has been received in the first request output step (S1030) An old request can be output (S1032). After outputting the request, the number of first tokens is decreased by one in the first token decreasing step (S1040), and the process may proceed to the first token check step (S1010) again.

Next, when proceeding to the second token checking step (S1050), it is possible to check whether a second token usable for outputting a request of the host apparatus corresponding to the selected request queue remains. If the number of remaining second tokens is greater than 0 in the second token checking step (S1050), the process proceeds to a second open row detecting step (S1060) (YES). If the number of remaining second tokens is 0, And the token initialization step (S1090).

In the second open row detection step S1060, if there is a request corresponding to the open row among the requests stored in the request queue, the process proceeds to YES and the second request output step S1070 And if the request corresponding to the open row of the requests stored in the request queue does not exist, the flow advances to 'NO' to proceed to the selection end and token initialization step (S1090). After outputting the request, the number of first tokens is decreased by one in the second token decreasing step (S1080), and the process may proceed to the second token checking step (S1050) again.

In the selection end and token initialization step S1090, the selection state of the currently selected request queue is terminated and the process proceeds to a step for selecting another request queue (see FIG. 9). At this time, the number of the first and second tokens corresponding to the currently selected request queue and the host device can be initialized.

In the step of outputting a request using the first token in the above-described request outputting process, the request is output in order according to the first-ready (FR-FCFS) rule, It is possible to output the requests in order according to a first-ready (FR) rule. In addition, when determining the order of the requests according to the FR-FCFS rule or the FR rule, it is possible to consider whether or not the row address of the memory device corresponding to the just selected and outputted request is the same.

11 is a flowchart illustrating a scheduling method of a memory controller according to an embodiment of the present invention. The scheduling method of the memory controller shown in FIG. 11 is different from the scheduling method of the memory controller shown in FIG. 9, and the scheduling of requests can be performed considering the request from the host device.

Referring to FIG. 11, a method of scheduling a memory controller includes a token allocation step S1110, a host device designation step S1120, a host device selection step S1130, a request output step S1140, a request reception step S1150, - Scheduling a host device request (S1160).

The scheduling method of the memory controller of FIG. 11 operates in a manner similar to the scheduling method of the memory controller of FIG. 9, except that in step S1150, The scheduling of the requests transmitted from the host apparatus is completed, and the scheduling of the requests transmitted from the host apparatus is completed (step S1160) (RESUME) the previous step. Also, if the request is not input from the host device, the process can proceed to 'NO' in 'S1101'.

At this time, the priority-host device request scheduling step (S1160) may be the same as the scheduling method of the requests stored in the selected request queue described in FIG.

12 is a flowchart illustrating a method of scheduling a memory controller according to an embodiment of the present invention. The scheduling method of the memory controller shown in FIG. 12 differs from the scheduling method of the memory controller shown in FIG. 9 in that the number of the first and second tokens allocated to each host device is calculated in consideration of the actual bandwidth and the target bandwidth of each host device Can be adjusted.

12, the scheduling method of the memory controller includes a token allocation step S1210, a host device designation step S1220, a host device selection step S1230, a request output step S1240, a request reception step S1250, A monitoring step S1260 and a bandwidth comparison step S1270.

In the case of the token allocation step S1210, an initial allocation step S1211, a token increase step S1212, and a token reduction step S1213 may be performed. The initial allocation step S1211 may be a step in which a token is allocated without comparing the actual bandwidth and the target bandwidth. The step of increasing the token S1212 may be a step of increasing the number of the first or second tokens allocated to the host device in which the actual bandwidth is larger than the target bandwidth as a result of the bandwidth monitoring. The token reduction step S1213 may be a step of increasing the number of first or second tokens allocated to the host apparatus whose actual bandwidth is smaller than the target bandwidth as a result of bandwidth monitoring.

If the token is not reassigned in 'S1201', the process can proceed to 'NO'. If the token is reallocated, the process proceeds to 'YES' to proceed with the bandwidth monitoring step S1260 and the bandwidth comparing step S1270.

In the bandwidth monitoring step S1260, the actual bandwidth corresponding to each host device can be detected by referring to the request scheduling result of the memory controller during the set interval. The bandwidth comparing step S1270 compares the actual bandwidth with the target bandwidth, and if the actual bandwidth is larger than the target bandwidth, proceeds to the step of increasing the token (S1212), compares the actual bandwidth with the target bandwidth, If it is smaller, the process may proceed to the token reduction step S1213.

The above-mentioned request may include a content for performing a write operation to store data transmitted from a host device in a memory device, or may be a content for outputting data from a memory device and performing a read operation to transfer the data to the host device.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. It will be apparent to those skilled in the art that various modifications and variations are possible in light of the above teachings.

Claims (20)

A plurality of request queues for storing requests transmitted from corresponding ones of a plurality of host devices;
A token information generator for generating information on the number of first and second tokens corresponding to the plurality of host devices, respectively; And
Outputting one request per one of said first tokens, outputting a request for said first token, and outputting a request for said second token, A request scheduler for outputting one request per one of the second tokens,
.
The method according to claim 1,
The token information generating unit
And to initialize information about the first and second number of tokens corresponding to a previously selected request queue if the next request queue of the selected request queue is selected.
The method according to claim 1,
The request scheduler
The first or second token is still present but there is no more request to output to the selected request queue or after using all of the first token and using the second token among the requests stored in the selected request queue And to select the next request queue of the selected request queue if there is no request.
The method according to claim 1,
The request scheduler
When a request is output using the first token, the request is output in order according to a first-ready (FR-FCFS) rule. When a request is output using the second token, first-ready rules to output requests in order.
5. The method of claim 4,
And when the order of the request is determined according to the FR-FCFS rule or the FR rule, it considers whether or not the row address is the same as the row address of the memory device corresponding to the selected request.
The method according to claim 1,
Wherein the number of the first and second tokens allocated to the plurality of host devices is determined according to characteristics of the host device.
The method according to claim 1,
The request scheduler
When one or more host devices of the plurality of host devices are set as a first-priority host device, when a request is received from the first-priority host device, outputting requests of the currently selected request queue is stopped, And outputting the requests.
The method according to claim 1,
The request scheduler
And skips the selection of the request queue from which the request is not stored among the plurality of request queues.
The method according to claim 1,
A bandwidth monitor for monitoring a bandwidth between the plurality of host devices and the memory controller; And
And a bandwidth control unit for controlling information on the number of the first and second tokens stored in the token information storage unit by referring to the monitoring result of the bandwidth monitoring unit,
Further comprising a memory controller.
10. The method of claim 9,
The bandwidth controller
Wherein the number of the first or second tokens allocated to the host device whose actual bandwidth is smaller than the target bandwidth is decreased and the number of the first or second tokens allocated to the host device 2 A memory controller that increases the number of tokens.
1. A request scheduling method for a memory controller that outputs requests sent from a plurality of host devices to a memory device,
Assigning one or more first and second tokens to each of the plurality of host devices;
Storing a request transmitted from the plurality of host devices in a request queue corresponding to each of the plurality of host devices;
Selecting a request queue of one of the plurality of request queues;
Outputting the requests stored in the selected request queue for each of the first tokens one by one; And
Outputting the requests stored in the selected request queue per one of the second tokens one by one if the first token is exhausted
And a memory controller for storing the request.
12. The method of claim 11,
Wherein when the next request queue of the selected request queue is selected, the number of the first and second tokens allocated to the previously selected request queue is initialized.
12. The method of claim 11,
The first or second token is still present but there is no more request to output to the selected request queue or after using all of the first token and using the second token among the requests stored in the selected request queue And if there is no request, selecting a next request queue of the selected request queue.
12. The method of claim 11,
In the step of outputting the request using the first token, the request is outputted in order according to the first-ready (FR-FCFS) rule,
And outputting a request using the second token, the request being output in order according to a first-ready (FR) rule.
15. The method of claim 14,
When determining the order of a request according to the FR-FCFS rule or the FR rule, whether or not the address is the same as the row address of the memory device corresponding to the selected request.
12. The method of claim 11,
In the step of allocating the first and second tokens
Wherein the number of the first and second tokens assigned to the plurality of host devices is determined according to the characteristics of the host device. 12. The method of claim 11,
Stopping operations of outputting requests of a currently selected request queue when the request is received from the first-priority host apparatus when one or more host apparatuses of the plurality of host apparatuses are set as a first-priority host apparatus, Selecting a request queue corresponding to the highest priority host device in the request queue.
12. The method of claim 11,
In the step of selecting the request queue
Wherein the plurality of request queues are repeatedly selected in a predetermined order, and a selection of a request queue in which the request is not stored is skipped.
12. The method of claim 11,
Monitoring the bandwidth between the plurality of host devices and the memory controller
Further comprising the steps of:
20. The method of claim 19,
In the step of allocating the first and second tokens
Wherein the monitoring unit monitors the monitoring result of the bandwidth monitoring unit to reduce the number of the first or second tokens allocated to the host apparatus whose actual bandwidth is smaller than the target bandwidth among the plurality of host apparatuses, And to increase the number of the first or second tokens assigned to the host device.

Images