(19)
(11)EP 3 537 281 A1

(12)EUROPEAN PATENT APPLICATION
published in accordance with Art. 153(4) EPC

(43)Date of publication:
11.09.2019 Bulletin 2019/37

(21)Application number: 17889889.6

(22)Date of filing:  28.10.2017
(51)Int. Cl.: 
G06F 3/06  (2006.01)
(86)International application number:
PCT/CN2017/108194
(87)International publication number:
WO 2018/126771 (12.07.2018 Gazette  2018/28)
(84)Designated Contracting States:
AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR
Designated Extension States:
BA ME
Designated Validation States:
MA MD

(30)Priority: 05.01.2017 CN 201710008824

(71)Applicant: Huawei Technologies Co., Ltd.
Longgang District Shenzhen, Guangdong 518129 (CN)

(72)Inventors:
  • YU, Si
    Shenzhen Guangdong 518129 (CN)
  • GONG, Junhui
    Shenzhen Guangdong 518129 (CN)
  • ZHAO, Cong
    Shenzhen Guangdong 518129 (CN)
  • WANG, Cheng
    Shenzhen Guangdong 518129 (CN)
  • LU, Yue
    Shenzhen Guangdong 518129 (CN)

(74)Representative: Kreuz, Georg Maria 
Huawei Technologies Duesseldorf GmbH Riesstraße 25
80992 München
80992 München (DE)

  


(54)STORAGE CONTROLLER AND IO REQUEST PROCESSING METHOD


(57) This application relates to the field of storage technologies, and discloses a storage controller. The storage controller includes a distribution core, a plurality of ranking cores, and a request processing core. The three types of cores are respectively configured to: distribute input/output IO requests to different ranking cores, generate a processing ranking index for each IO request, and process the IO request based on a value of the processing ranking index of the IO request, to flexibly schedule an IO request received by the storage controller.




Description

TECHNICAL FIELD



[0001] This application relates to the field of storage technologies, and in particular, to a storage controller and an input/output (English full name: input/output, IO for short) request processing method performed by the storage controller.

BACKGROUND



[0002] As shown in FIG. 1, a storage array is usually used in a large-scale storage scenario, and includes a plurality of storage mediums and a storage controller. The storage medium may include a hard disk drive (English full name: hard disk drive, HDD for short) and a solid state drive (English full name: solid state drive, SSD for short). A client sends an IO request to the storage controller by using a communications network. The storage controller processes the received IO request. For example, if the IO request is a read request, the storage controller determines a storage medium to which the read request points. Then, the storage controller reads corresponding data from the one or more storage mediums, and returns the data to the client.

[0003] The storage controller virtualizes the storage mediums in the storage array into a plurality of storage units. The IO request received by the storage controller usually points to a storage unit. If different storage types are used, the storage controller virtualizes the plurality of storage mediums into different types of storage units (English: storage unit). For example, if block storage is used, the storage controller virtualizes the plurality of storage mediums into one or more logical unit numbers (English full name: logical unit number, LUN for short). Each IO request of the client points to a LUN. If file storage is used, each IO request of the client points to a file system. If object (English: object) storage is used, each IO request of the client points to a bucket (English: bucket).

[0004] Because of service requirements, a user usually needs to set IO per second (English: input/output per second, IOPS for short) parameters for different storage units. If the client sends a relatively large quantity of IO requests, these IO requests point to different storage units. Because the storage controller processes an IO request at a limited speed, the storage controller needs to schedule received IO requests to fulfill QoS parameters of the plurality of storage units as much as possible.

[0005] An IOPS parameter fulfillment rate of an existing IO request scheduling method is relatively low.

SUMMARY



[0006] This application provides a storage controller, to improve an IOPS fulfillment rate.

[0007] A first aspect of this application provides a storage controller. The storage controller is applicable to a storage system having a plurality of storage units. The storage controller includes a memory device and a plurality of cores. The plurality of cores include at least one distribution core, a plurality of ranking cores, and at least one request processing core. The memory device further stores a plurality of input/output IO requests, each IO request points to one storage unit, and a corresponding shared processing ranking index is further set for each storage unit in the memory device.

[0008] During operation, the distribution core executes code stored in the memory device to: receive an IO request stored in the memory device, and distribute the received IO request to the plurality of ranking cores.

[0009] During operation, each ranking core executes code stored in the memory device to perform the following actions: obtaining an IO request for which a processing ranking index is to be generated and that is distributed by each distribution core; determining a target storage unit to which the IO request for which a processing ranking index is to be generated points; obtaining an IO per second parameter of the target storage unit; generating, based on the IO per second parameter of the target storage unit and a value of a shared processing ranking index corresponding to the target storage unit, the processing ranking index for the IO request for which a processing ranking index is to be generated; updating, by using the processing ranking index of the IO request for which a processing ranking index is to be generated, the shared processing ranking index corresponding to the target storage unit; and storing, in an index queue corresponding to the ranking core, the processing ranking index of the IO request for which a processing ranking index is to be generated, where the index queue corresponding to said each ranking core is stored in the memory device and includes processing ranking indexes that are generated by said each ranking core for the IO requests pointing to the plurality of storage units.

[0010] During operation, the request processing core executes code stored in the memory device to periodically process an IO request corresponding to a smallest processing ranking index in the index queue corresponding to said each ranking core.

[0011] The distribution core, the plurality of ranking cores, and the request processing core may work in parallel.

[0012] The storage controller generates a processing ranking index for each IO request, and determines a processing order based on values of the processing ranking index of the IO requests, thereby effectively improving an IOPS parameter fulfillment rate.

[0013] In addition, when generating the processing ranking index for the IO request, each ranking core does not need to access another ranking core to obtain a processing ranking index generated by the another ranking core for the IO request, thereby improving processing efficiency.

[0014] Each ranking core described according to any aspect of this application or any implementation of any aspect is any one of the plurality of ranking cores.

[0015] The target storage unit described according to any aspect of this application or any implementation of any aspect is a logical unit number LUN, a file system, or a bucket.

[0016] With reference to the first aspect, in a first implementation of the first aspect, said each ranking core generates the processing ranking index for the IO request for which a processing ranking index is to be generated by performing the following operation:
calculating, based on a sum of the value of the shared processing ranking index corresponding to the target storage unit and a ratio of K to the IO per second parameter of the target storage unit, the processing ranking index of the IO request for which a processing ranking index is to be generated, where K is a positive number.

[0017] With reference to the first implementation of the first aspect, in a second implementation of the first aspect, before calculating the processing ranking index of the IO request for which a processing ranking index is to be generated, said each ranking core is further configured to obtain a current system time.

[0018] Therefore, said each ranking core calculates the processing ranking index of the IO request for which a processing ranking index is to be generated by performing the following operation:
using, as the processing ranking index of the IO request for which a processing ranking index is to be generated, a larger one of the current system time and the sum of the value of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit.

[0019] The system time is considered during the calculation of the processing ranking index, thereby improving IO request scheduling accuracy.

[0020] With reference to the first implementation or the second implementation of the first aspect, in a third implementation of the first aspect, said each ranking core is further configured to: at a moment after which the processing ranking index is generated for the IO request for which a processing ranking index is to be generated, determine that there is no IO request that has been distributed to said each ranking core, that points to the target storage unit, and for which no index has been generated.

[0021] Subsequently, said each ranking core calculates a sum, at the moment, of the value of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit, to use the sum as a pending processing ranking index. In addition, said each ranking core stores the pending ranking index in the index queue corresponding to said each ranking core. During operation of the ranking cores, once it is determined that there is no IO request pointing to the target storage unit in the IO requests that have been distributed to the ranking cores and for which no index has been generated, the pending processing ranking index is generated.

[0022] With reference to the third implementation of the first aspect, in a fourth implementation of the first aspect, within duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core, an IO request corresponding to a processing ranking index that is greater than the pending processing ranking index and that is included in the index queue corresponding to said each ranking core cannot be processed by the request processing core.

[0023] Said each ranking core is further configured to remove the pending processing ranking index from the index queue corresponding to said each ranking core if an IO request pointing to the target storage unit is distributed to said each ranking core after the moment or the duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core exceeds a preset threshold.

[0024] Application of the pending processing ranking index can improve IO request scheduling accuracy.

[0025] With reference to any one of the first aspect or the implementations of the first aspect, in a sixth implementation of the first aspect, the request processing core periodically processes the IO request corresponding to the smallest processing ranking index in the index queue corresponding to said each ranking core by performing the following operations: periodically accessing the index queue corresponding to said each ranking core; and processing an IO request, during each access, corresponding to the smallest processing ranking index in the index queue corresponding to said each ranking core.

[0026] A second aspect of this application provides an IO request processing method. During operation, the storage controller according to the first aspect performs the method. The method includes: receiving, by the distribution core, an IO request, and distributing the received IO request to the plurality of ranking cores; obtaining, by each ranking core, an IO request for which a processing ranking index is to be generated and that is distributed by each distribution core; determining, by said each ranking core, a target storage unit to which the IO request for which a processing ranking index is to be generated points; obtaining, by said each ranking core, an IO per second parameter of the target storage unit; generating, by said each ranking core based on the IO per second parameter of the target storage unit and a shared processing ranking index corresponding to the target storage unit, the processing ranking index for the IO request for which a processing ranking index is to be generated; updating, by said each ranking core by using the processing ranking index of the IO request for which a processing ranking index is to be generated, the shared processing ranking index corresponding to the target storage unit; storing, by said each ranking core in an index queue corresponding to said each ranking core, the processing ranking index of the IO request for which a processing ranking index is to be generated, where the index queue corresponding to said each ranking core is stored in the memory device and includes processing ranking indexes that are generated by said each ranking core for the IO requests pointing to the plurality of storage units; and periodically processing, by the request processing core, an IO request corresponding to a smallest processing ranking index in the index queue corresponding to said each ranking core.

[0027] With reference to the second aspect, in a first implementation of the second aspect, the generating, by said each ranking core based on the IO per second parameter of the target storage unit and a shared processing ranking index corresponding to the target storage unit, the processing ranking index for the IO request for which a processing ranking index is to be generated includes: calculating, based on a sum of the shared processing ranking index corresponding to the target storage unit and a ratio of K to the IO per second parameter of the target storage unit, the processing ranking index of the IO request for which a processing ranking index is to be generated, where K is a positive number.

[0028] With reference to the first implementation of the second aspect, in a second implementation of the second aspect, before the calculating, by said each ranking core, the processing ranking index of the IO request for which a processing ranking index is to be generated, the method further includes: obtaining, by said each ranking core, a current system time; and the calculating, by said each ranking core based on a sum of the shared processing ranking index corresponding to the target storage unit and a ratio of K to the IO per second parameter of the target storage unit, the processing ranking index of the IO request for which a processing ranking index is to be generated includes: using, by said each ranking core, as the processing ranking index of the IO request for which a processing ranking index is to be generated, a larger one of the current system time and the sum of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit.

[0029] With reference to the first implementation and the second implementation of the second aspect, in a third implementation of the second aspect, the method further includes: determining, by said each ranking core, at a moment after which the processing ranking index is generated for the IO request for which a processing ranking index is to be generated, that there is no IO request that has been distributed to said each ranking core, that points to the target storage unit, and for which no index has been generated; calculating, by said each ranking core, a sum, at the moment, of the value of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit, to use the sum as a pending processing ranking index; and storing, by said each ranking core, the pending processing ranking index in the index queue corresponding to said each ranking core.

[0030] With reference to the third implementation of the second aspect, in a fourth implementation of the second aspect, within duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core, an IO request corresponding to a processing ranking index that is greater than the pending processing ranking index and that is included in the index queue corresponding to said each ranking core cannot be processed by the request processing core; and said each ranking core removes the pending processing ranking index from the index queue corresponding to said each ranking core if an IO request pointing to the target storage unit is distributed to said each ranking core after the moment or the duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core exceeds a preset threshold.

[0031] With reference to any one of the second aspect or the implementations of the second aspect, in a fifth implementation of the second aspect, the periodically processing, by each request processing core, an IO request corresponding to a smallest processing ranking index in the index queue corresponding to said each ranking core includes:

periodically accessing the index queue corresponding to said each ranking core; and

processing an IO request, during each access, corresponding to the smallest processing ranking index in the index queue corresponding to said each ranking core.



[0032] A third aspect of this application provides a storage medium. The storage medium stores program code. When the program code is run by a storage controller, the storage controller performs the IO request processing method according to any one of the second aspect or the implementations of the second aspect. The storage medium includes but is not limited to a read-only memory, a random access memory, a flash memory, an HDD, or an SSD.

[0033] A fourth aspect of this application provides a computer program product. The computer program product includes program code. When the computer program product is executed by a storage controller, the storage controller performs the IO request processing method according to any one of the second aspect or the implementations of the second aspect. The computer program product may be a software installation package. If the IO request processing method according to any one of the second aspect or the implementations of the second aspect, the computer program product may be downloaded to the storage controller, and the computer program product runs on the storage controller.

BRIEF DESCRIPTION OF DRAWINGS



[0034] 

FIG. 1 is a schematic structural hierarchical diagram of a storage system according to an embodiment of this application;

FIG. 2 is a schematic structural hierarchical diagram of a storage controller;

FIG. 3 is a schematic structural diagram of a memory device;

FIG. 4 is a schematic structural diagram of another memory device;

FIG. 5 is a schematic diagram of a process of generating processing ranking indexes;

FIG. 6 is a schematic diagram of another process of generating processing ranking indexes; and

FIG. 7-1 to FIG. 7-5 are schematic diagrams of another process of generating processing ranking indexes.


DESCRIPTION OF EMBODIMENTS



[0035] The following describes the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application.

[0036] Throughout this specification, a processor includes one or more central processing units (English full name: central processing unit), and each central processing unit includes one or more cores (English: core).

[0037] Throughout this specification, a storage unit may be a LUN, a file system, or a bucket, respectively corresponding to a case in which block storage, file storage, or object storage is used in a storage system. For example, the storage system in this specification presents P LUNs for a client, and P is a positive integer greater than 1.

[0038] Throughout this specification, an IO request includes IO data and metadata. The IO data includes information such as to-be-operated data of the IO request and an address of the to-be-operated data. The metadata includes a target storage unit ID of the IO request. The target storage unit ID may be a LUN ID, a file system ID, or a bucket ID.

[0039] Throughout this specification, a function Max {x, y} is used to return a larger value of x andy.

[0040] Throughout this specification, an IOPS parameter may be an IOPS of a storage unit, or an IOPS processing weight of a storage unit. The IOPS processing weight refers to a proportion of resources used by a storage array to process an IO request pointing to each storage unit. Therefore, the IOPS parameter may be set by a user according to a service requirement. For example, the user determines, according to a service requirement, a minimum of IOPS of a storage unit related to a service, or the user determines a weight of resources of the storage array that need to be occupied by an IO request of a storage unit related to a service. The IOPS parameter may alternatively be set according to a level of a user. For example, an advanced user has a relatively large IOPS parameter, to ensure experience of the advanced user. A storage controller stores IOPS parameters of different storage units.

[0041] An architecture of a storage controller applied to this embodiment of this application

[0042] As shown in FIG. 2, a processor of the storage controller includes a plurality of cores, a memory device, and a communications interface. Each core establishes a communication connection to the memory device. The storage controller communicates with a client and a storage medium by using the communications interface. An IO request obtained from the communications interface is stored in an IO storage space of the memory device.

[0043] During IO request scheduling and processing, there are mainly three types of programs: an IO request distribution program, an IO request ranking program, and an IO request processing program being used. The IO request distribution program, the IO request ranking program, and the IO request processing program each are implemented by a core running code in the memory device. A core running the IO request distribution program is referred to as a distribution core, a core running the IO request ranking program is referred to as a ranking core, and a core running an IO request processing program is referred to as a request processing core.

[0044] A quantity of cores to be respectively allocated to the three types of programs may be adjusted depending on running pressure of the three types of programs. A core on which each program runs may be adjusted depending on a load status of each core. In FIG. 2, for example, a core 1 is configured to execute the IO request distribution program, a core 2 to a core n are configured to execute the IO request ranking program, a core n+1 to a core n+m are configured to execute the IO request processing program, and a core n+m+1 is configured to execute an operating system of the storage controller.

[0045] The IO request distribution program distributes IO requests in the IO storage space, and distributes the IO requests to subspaces of the cores running the IO request ranking program. In FIG. 2, for example, subspaces of the core 2 to the core n are located inside a space of the core 1. In practice, the subspaces of the core 2 to the core n may be alternatively located outside the space of the core 1, or respectively located inside spaces of the cores.

[0046] In the process in which the IO request distribution program distributes the IO requests, subsequent load balance of each IO request ranking program is mainly considered, and distribution of all IO requests pointing to a LUN to a space of a core is not considered. For example, the IO request distribution program may distribute a plurality of received IO requests to each IO request ranking program in turn, to ensure that a same quantity of IO requests are distributed to all the IO request ranking programs.

[0047] Subsequently, each IO request ranking program reads IO requests distributed to the IO request ranking program and sorts the IO requests. A sorting result is stored in an index queue in a space of a core in which each IO request ranking program is located. The index queue may be implemented by using different data structures, such as a pile (English: pile) and a first-in-first-out queue. Each IO request ranking program generates one processing ranking index for each IO request in a subspace of the IO request ranking program, and then sorts the processing ranking indexes of the IO requests in the index queue. An IO request with small index is preferentially to be processed by an idle IO request processing program.

[0048] The IO request processing program may specifically perform, according to a type of an IO request, a write operation or a read operation corresponding to the IO request. The IO request processing program may alternatively be configured to perform configuration, deduplication, or the like on data carried by the IO request.

[0049] For example, block storage is used in a storage array in which the storage controller is located and a storage medium of the storage array is virtualized into 100 LUNs. Because of service requirements, IOPS parameters need to be set for some LUNs. Each IO request received by the storage controller points to a LUN. A client generates an IO request at an indefinite speed. Therefore, a quantity of IO requests that point to different LUNs and that are generated per second may vary greatly. Because processing efficiency of the IO request processing program is limited, the sorting result of the IO request ranking program affects whether an IOPS parameter of each LUN can be fulfilled. For example, an IOPS parameter of a LUN 1 is 1000, and an IOPS parameter of a LUN 2 is 200. However, because a relatively large quantity of IO requests pointing to the LUN 2 are generated within a period of time, 1000 IO requests pointing to the LUN 1 and 2000 10 requests pointing to the LUN 2 are stored in the IO storage space of the memory device at a moment. The 3000 IO requests are distributed to IO request ranking programs in the core 2 to the core n+1 for sorting. If the IO request ranking programs schedule the IO requests based only on the IOPS parameter of the LUN 1 and the IOPS parameter of the LUN 2, finally, it is generally difficult for the n IO request ranking programs to fulfill the IOPS parameter of the LUN 1 and the IOPS parameter of the LUN 2. If the IO request ranking programs communicate with each other during generation of processing ranking indexes, this is helpful in fulfilling a lower limit of IOPS of each LUN. However, overheads of the communication between the IO request ranking programs are very high.

[0050] This application provides an IO request processing method. The method is applicable to the storage controller shown in FIG. 2.

[0051] The memory device of the storage controller shown in FIG. 2 maintains one shared processing ranking index for each storage unit. The P shared processing ranking indexes may be read or written by each IO scheduler. Initial values of the shared processing ranking indexes are the same. For example, the initial values of the shared processing ranking indexes may be 0.

[0052] In practice, there may be a plurality of shared processing ranking index implementations. For example, the shared processing ranking indexes may be combined into a table, and set in the storage space of the memory device. All the shared processing ranking indexes are created by the operating system before the storage controller starts to distribute IO description information.

[0053] As shown in FIG. 3, the communications interface receives a plurality of IO requests sent by the client, and stores the plurality of IO requests in the IO storage space.

[0054] The IO request distribution program generates IO description information for each IO request, and establishes a mapping relationship between each IO request and the IO description information of the IO request. The IO description information of each IO request includes a LUN ID carried in metadata of the IO request.

[0055] Because the IO requests occupy a relatively large space, IO description information may be generated for each IO request during sorting of the IO requests. Subsequently, the IO request ranking program generates a processing ranking index based on the IO description information, to reduce read/write load on the memory device.

[0056] As shown in FIG. 4, the IO request distribution program distributes the plurality of pieces of IO description information to subspaces of cores in which the IO request ranking programs are located.

[0057] The IO request distribution program may create, in a subspace of a core in which each IO request ranking program is located, a queue for each LUN, and separately store, in the queue of each LUN, IO description information allocated to the subspace of the core in which the IO request ranking program is located, so that in a subsequent step, the IO request ranking program identifies a LUN to which each piece of IO description information points.

[0058] How an IO request ranking program running on the core 2 generates a processing ranking index for a piece of IO description information is described below by using FIG. 5. During operation, all IO request ranking programs generate a processing ranking index for each piece of IO description information by using a same method.

[0059] In FIG. 5, IO description information A-B-C indicates Cth IO description information that points to a LUN B and that is distributed to an IO request ranking program running on a core A. Correspondingly, a processing ranking index A-B-C indicates a processing ranking index of the IO description information A-B-C.

[0060] For example, the IO request ranking program running on the core 2 currently generates a processing ranking index for IO description information 2-1-3. Therefore, processing ranking indexes of IO description information 2-1-1 and IO description information 2-1-2 have been stored by the IO request ranking program in an index queue of the core 2.

[0061] The IO request ranking program running on the core 2 obtains the IO description information 2-1-3 from a subspace of the core 2, and obtains a LUN ID corresponding to the IO description information 2-1-3.

[0062] The IO request ranking program running on the core 2 obtains an IOPS parameter of a LUN 1 according to the LUN ID.

[0063] The IO request ranking program running on the core 2 obtains, according to the LUN ID, a value of a shared processing ranking index corresponding to the LUN 1, that is, a value of a shared processing ranking index 1.

[0064] The IO request ranking program running on the core 2 obtains a current system time by using an operating system interface.

[0065] The current system time may specifically refer to a quantity of nanoseconds that have elapsed from startup of the storage controller to invocation of the operating system interface.

[0066] The IO request ranking program running on the core 2 calculates a processing ranking index 2-1-3 of the IO description information 2-1-3.

[0067] The processing ranking index 2-1-3 = Max {the value of the shared processing ranking index 1 + K / the IOPS parameter of the LUN 1, the current system time}. K is a positive number, and commonly, a value of K is 1.

[0068] The IO request ranking program running on the core 2 updates the shared processing ranking index corresponding to the LUN 1 by using the processing ranking index 2-1-3.

[0069] After generating a processing ranking index of IO description information corresponding to a LUN, each IO request ranking program updates a shared processing ranking index corresponding to the LUN by using the generated processing ranking index. Therefore, the P shared counters respectively record processing ranking indexes of IO requests that point to each LUN and that are the most recently processed. Therefore, when any IO request ranking program calculates a processing ranking index of IO description information pointing to a LUN p, a value of a shared processing ranking index p is equal to a processing ranking index of a previous piece of IO description information pointing to the LUN p. The LUN p is any one of the P LUNs presented by the storage system.

[0070] If the IO request ranking program running on the core 2 currently generates a processing ranking index for a processing ranking index 2-1-1, the processing ranking index 2-1-1 = Max {an initial processing ranking index + K / the IOPS parameter of the LUN 1, a current system time}. The initial processing ranking index may be 0.

[0071] The IO request ranking program running on the core 2 stores the processing ranking index 2-1-3 in the index queue of the core 2.

[0072] A correspondence is established between the IO description information 2-1-3 and the processing ranking index 2-1-3, or a correspondence is established between the processing ranking index 2-1-3 and an IO request for which the IO description information 2-1-3 is generated, so that the IO request corresponding to the processing ranking index 2-1-3 can be determined in a subsequent step.

[0073] By using the foregoing steps, each IO request ranking program generates a processing ranking index for each piece of IO description information distributed to the IO request ranking program, and stores the processing ranking index in an index queue of the IO request ranking program. Therefore, processing ranking indexes of one or more IO requests that are not processed are stored in index queues of the core 2 to the core n.

[0074] After an IO request processing program running on any core processes an IO request, the operating system learns that the IO request processing program enters an idle state.

[0075] The operating system records a processing order of each IO request processing program, that is, the processing order indicates an IO request corresponding to a smallest processing ranking index in an index queue that is to be subsequently processed by an IO request processing program after entering an idle state. To ensure fulfillment of an IOPS parameter, the processing order needs to enable an IO request processing program to process processing ranking indexes in all the index queues at same or close frequency. To be specific, an IO request processing program periodically processes an IO request corresponding to a smallest processing ranking index in each index queue.

[0076] The processing order may be: each IO request processing program polls all the index queues in an order of the core 2 to the core n, and processes an IO request corresponding to a smallest processing ranking index in an index queue accessed each time. Every time after processing an IO request corresponding to a smallest processing ranking index in an index queue, an idle IO request processing program processes an IO request corresponding to a smallest processing ranking index in a next index queue.

[0077] Alternatively, if m=n-1, that is, a quantity of IO request ranking programs is the same as that of IO request processing programs, the operating system binds the IO request processing programs to the index queues respectively one by one. In this case, after an IO request processing program enters an idle state, the operating system determines that the idle IO request processing program subsequently processes an IO request corresponding to a smallest processing ranking index in an index queue corresponding to the idle IO request processing program.

[0078] After the operating system determines an index queue in which an IO request corresponding to a smallest processing ranking index is to be processed by the idle IO request processing program, the operating system selects the smallest processing ranking index from the index queue, and instructs the idle IO request processing program to process the IO request corresponding to the smallest processing ranking index, or the operating system instructs the idle IO request processing program to access the index queue, and process the IO request corresponding to the smallest processing ranking index in the index queue.

[0079] An order in which some of the foregoing steps are performed may be adjusted. Both the step of obtaining the IOPS parameter of the LUN 1 and the step of obtaining a current system time may be performed at any moment before the processing ranking index 2-1-3 is generated.

[0080] The foregoing step of obtaining a system time is an optional step. When the step is not performed, the value of the generated processing ranking index 2-1-3 = the value of the shared processing ranking index 1 + K / the IOPS parameter of the LUN 1.

[0081] The idle IO request processing program preferentially processes the IO request corresponding to the smallest processing ranking index in each index queue. Therefore, if no IO request pointing to a LUN is distributed within a period of time to the IO request ranking program running on the core 2, but IO requests pointing to another LUN are continuously distributed within this period of time to the IO request ranking program running on the core 2, in a subsequent period of time, once IO requests pointing to the LUN are distributed to the IO request ranking program running on the core 2, all processing ranking indexes of these IO requests pointing to the LUN may be smaller than processing ranking indexes of the IO requests pointing to the another LUN. Consequently, the IO requests pointing to the LUN are continuously preferentially processed by the idle IO request processing program, causing starvation of the IO requests pointing to the another LUN. Therefore, the consideration of the system time in the calculation of the processing ranking index prevents, if there is a long idle LUN, the IO requests pointing to the another LUN from subsequently being blocked after the IO requests pointing to the LUN are distributed to the subspace of the IO request ranking program running on the core 2, improving IO request scheduling accuracy.

[0082] For example, at a moment 1, processing ranking indexes recorded in the index queue of the core 2 include:

a processing ranking index 2-1-1 = 3 a processing ranking index 2-2-1 = 2.2

a processing ranking index 2-1-2 = 3.5 a processing ranking index 2-2-2 = 2.8

a processing ranking index 2-1-3 = 5.5 a processing ranking index 2-2-3 = 3.0



[0083] At a moment 2, 10 requests corresponding to processing ranking indexes that have been stored in the index queue at the moment 1 have been processed. New processing ranking indexes generated between the moment 1 and the moment 2 include:

a processing ranking index 2-1-4 = 6

a processing ranking index 2-1-5 = 7.5

a processing ranking index 2-1-6 = 9.5

a processing ranking index 2-1-7 = 10.5

a processing ranking index 2-1-8 = 12



[0084] That is, within duration from the moment 1 to the moment 2, no new IO description information pointing to the LUN 2 is distributed to the IO request ranking program running on the core 2. Therefore, after the moment 2, if the system time is not considered in the calculation of the processing ranking index, and if new IO description information pointing to the LUN 2 is distributed to the IO request ranking program running on the core 2, processing ranking indexes of these pieces of newly distributed IO description information pointing to the LUN 2 are much smaller than a processing ranking index of IO description information pointing to the LUN 1. Consequently, when accessing the index queue of the core 2, the idle IO request processing program continuously processes newly distributed IO requests pointing to the LUN 2. However, if the system time is considered in the calculation of the processing ranking index, the processing ranking indexes of these pieces of newly distributed IO description information pointing to the LUN 2 may be equal to the current system time and not be much smaller than the processing ranking index of the IO description information pointing to the LUN 1.

[0085] Therefore, the consideration of the system time in the calculation of the processing ranking index prevents, if there is a long idle LUN within a period of time, IO requests pointing to another LUN from subsequently being blocked after IO requests pointing to the idle LUN arrive, improving IO request scheduling accuracy.

[0086] In the process of performing the foregoing steps, if the IO request ranking program running on the core 2 determines, at a moment, that the IO description information pointing to the LUN 1 in the index queue of the core 2 has been processed, as shown in FIG. 6, and after the processing ranking index 2-1-5 is generated, if there is no IO description information that points to the LUN 1 and for which no processing ranking index is generated in the IO description information that is distributed to the IO request ranking program running on the core 2, the IO request running on the core 2 generates a pending processing ranking index and stores the pending processing ranking index in the index queue of the core 2.

[0087] The pending processing ranking index = the value of the shared processing ranking index corresponding to the LUN 1 at the moment + K / the IOPS parameter of the LUN 1.

[0088] The pending processing ranking index is removed in one of the following two cases. In one case, new IO description information pointing to the LUN 1 is distributed to the IO request ranking program running on the core 2. In the other case, existence duration of the pending processing ranking index exceeds a preset threshold.

[0089] The step of generating a pending processing ranking index is an optional step. Processing ranking indexes in each index queue are sorted together with the pending processing ranking index. If the operating system determines, in the process of selecting an IO request to be subsequently processed for the idle IO request processing program, that a smallest processing ranking index in a current index queue is a pending processing ranking index, because the pending processing ranking index does not correspond to any IO request, the idle IO request processing program cannot process an IO request corresponding to a processing ranking index in the index queue. The operating system may reselect an index queue for the idle IO request processing program.

[0090] For example, the idle IO request processing program polls the index queues of the cores in an order of the core 2 to the core n. If the idle IO request processing program currently polls an index queue of a core 3, but finds that a smallest processing ranking index in the index queue of the core 3 is a pending processing ranking index, the idle IO request processing program skips the index queue of the core 3, and accesses an index queue of a core 4.

[0091] Different quantities of pieces of IO description information pointing to different LUNs are distributed to the IO request ranking programs, and a speed at which an IO request ranking program generates a processing ranking index for an IO request is usually much higher than a speed at which an IO request processing program processes an IO request. Therefore, if no pending processing ranking index is generated and a small quantity of pieces of IO description information pointing to the LUN 2 is distributed to the IO request ranking program running on the core 2, IO description information corresponding to a smallest processing ranking index in the index queue of the core 2 may be caused to point to the LUN 1 all the time, causing the idle IO request processing program to constantly process the IO request pointing to the LUN 1, and finally causing it difficult to fulfill the IOPS parameter of the LUN 2.

[0092] For example, at a moment 1, processing ranking indexes stored in the index queue of the core 2 include:

a processing ranking index 2-1-1 = 3 a processing ranking index 2-2-1 = 2.2

a processing ranking index 2-1-2 = 3.5 a processing ranking index 2-2-2 = 2.8

a processing ranking index 2-1-3 = 5.5 a processing ranking index 2-2-3 = 3.0



[0093] At a moment 2, 10 requests corresponding to some processing ranking indexes that have been generated at the moment 1 have been processed, and within duration from the moment 1 to the moment 2, new processing ranking indexes are generated. Therefore, processing ranking indexes recorded in the index queue of the core 2 include:

a processing ranking index 2-1-4 = 6 a processing ranking index 2-2-4 = 3.4

a processing ranking index 2-1-5 = 7.5

a processing ranking index 2-1-6 = 9.5



[0094] If in a subsequent period of time, no IO description information pointing to the LUN 2 is distributed to the IO request ranking program running on the core 2, but the IO description information pointing to the LUN 1 is constantly distributed to the IO request ranking program running on the core 2, and after an IO request corresponding to the processing ranking index 2-2-4 is processed, if no pending processing ranking index is generated, once accessing the index queue of the core 2, the idle IO request processing program constantly processes the IO request pointing to the LUN 1. Consequently, the IOPS parameter of the LUN 2 cannot be fulfilled. In contrast, a pending processing ranking index 2-2-6 is generated, and before the pending processing ranking index 2-2-6 is removed, a processing ranking index in the index queue of the core 2 that is greater than the pending processing ranking index 2-2-6 cannot be processed by the idle IO request processing program. Consequently, the idle IO request processing program needs to access another index queue. Therefore, use of the pending processing ranking index can improve IO request scheduling accuracy, thereby improving an IOPS parameter fulfillment rate.

[0095] For example, the processor includes three ranking cores, FIG. 7-1 is an initial state, and initial values of the shared processing ranking index 1 and the shared processing ranking index 2 are both 0. In FIG. 7-1, IO description information a-b-c indicates cth IO description information that points to a LUN b and that is distributed to a core a. Initial processing ranking indexes of IO description information pointing to a LUN 1 and IO description information pointing to a LUN 2 are both 0. An IOPS parameter of the LUN 1 is 1000, an IOPS parameter of the LUN 2 is 500, and K=1.

[0096] At a moment T1, an IO request ranking program running on a core 2 calculates a processing ranking index for IO description information 2-1-1, and the processing ranking index of the IO description information 2-1-1 is 0.001. Subsequently, the value of the shared processing ranking index 1 is updated to 0.001, as shown in FIG. 7-2.

[0097] At a moment T2, an IO request ranking program running on a core 3 calculates a processing ranking index for IO description information 3-1-1, where the processing ranking index is 0.002. Subsequently, the value of the shared processing ranking index 1 is updated to 0.002, as shown in FIG. 7-3.

[0098] At a moment T3, an IO request ranking program running on a core 4 calculates a processing ranking index for IO description information 4-1-1, where the processing ranking index is 0.003. Subsequently, the value of the shared processing ranking index 1 is updated to 0.003.

[0099] At a moment T4, the IO request ranking program running on the core 4 calculates a processing ranking index for IO description information 4-2-1, and the processing ranking index of the IO description information 4-2-1 is 0.002. Subsequently, the value of the shared processing ranking index 2 is updated to 0.002, as shown in FIG. 7-4.

[0100] At a moment T5, the IO request ranking program running on the core 2 calculates a processing ranking index for IO description information 2-1-2, where the processing ranking index is 0.004. Subsequently, the value of the shared processing ranking index 1 is updated to 0.004.

[0101] At a moment T6, the IO request ranking program running on the core 2 calculates a processing ranking index for IO description information 2-2-1, and the processing ranking index of the IO description information 2-2-1 is 0.004. Subsequently, the value of the shared processing ranking index 2 is updated to 0.004, as shown in FIG. 7-5.

[0102] Processes of subsequently generating a processing ranking index in the core 2, the core 3, and the core 4 are performed by analogy.

[0103] In the examples corresponding to FIG. 7-1 to FIG. 7-5, impact of a system time on a generation process of a processing ranking index is not considered.

[0104] As shown in FIG. 2, the storage controller applied to this application includes a bus, a processor, a memory device, and a communications interface. The processor, the memory device, and the communications interface communicate with each other by using the bus.

[0105] The memory device may include a volatile memory (English: volatile memory), such as a random access memory (English: random access memory, RAM for short).

[0106] The communications interface includes a network interface and a storage medium access interface, respectively configured to obtain an IO request sent by a client and access a storage medium.

[0107] The memory device stores code required for executing an IO request distribution program, an IO request ranking program, an IO request processing program, and an operating system. During operation of the storage controller, each core in the processor invokes code stored in the memory device, to perform the IO request processing method provided above.

[0108] In the foregoing embodiments, the descriptions of the embodiments have respective focuses. For a part that is not described in detail in an embodiment, refer to related descriptions in other embodiments.

[0109] The methods described with reference to the content disclosed in this application may be implemented in a manner of executing a software instruction by a processor. The software instruction may include a corresponding software module. The software module may be stored in a RAM, a flash memory, a read-only memory (English: read only memory, ROM for short), an erasable programmable read-only memory (English: erasable programmable read only memory, EPROM for short), an electrically erasable programmable read-only memory (English: electrically erasable programmable read only memory, EEPROM for short), a hard disk, a compact disc, or a storage medium in any other forms well-known in the art.

[0110] A person of skill in the art should be aware that in one or more of the foregoing examples, the functions described in this application may be implemented by using hardware or software. When this application is implemented by software, these functions may be stored in a computer-readable medium or transmitted as one or more instructions or code in the computer-readable medium. The storage medium may be any available medium accessible to a general or dedicated computer.

[0111] The objectives, technical solutions, and benefits of this application are further described in detail in the foregoing specific embodiments. It should be understood that the foregoing descriptions are merely specific embodiments of this application, but are not intended to limit the protection scope of this application. Any modification or improvement made based on the technical solutions of this application shall fall within the protection scope of this application.


Claims

1. A storage controller, wherein the storage controller is applicable to a storage system having a plurality of storage units, and comprises at least one distribution core, a plurality of ranking cores, at least one request processing core, and a memory device, the memory device stores a plurality of input/output IO requests, each IO request points to one storage unit, and a corresponding shared processing ranking index is further set for each storage unit in the memory device;
each distribution core is configured to: receive an IO request, and distribute the received IO request to the plurality of ranking cores;
each ranking core is configured to:

obtain an IO request for which a processing ranking index is to be generated and that is distributed by each distribution core;

determine a target storage unit to which the IO request for which a processing ranking index is to be generated points;

obtain an IO per second parameter of the target storage unit;

generate, based on the IO per second parameter of the target storage unit and a value of a shared processing ranking index corresponding to the target storage unit, the processing ranking index for the IO request for which a processing ranking index is to be generated;

update, by using the processing ranking index of the IO request for which a processing ranking index is to be generated, the shared processing ranking index corresponding to the target storage unit; and

store, in an index queue corresponding to said each ranking core, the processing ranking index of the IO request for which a processing ranking index is to be generated, wherein the index queue corresponding to said each ranking core is stored in the memory device and comprises processing ranking indexes that are generated by said each ranking core for the IO requests pointing to the plurality of storage units; and

each request processing core is configured to periodically process an IO request corresponding to a smallest processing ranking index in the index queue corresponding to said each ranking core.


 
2. The storage controller according to claim 1, wherein said each ranking core generates the processing ranking index for the IO request for which a processing ranking index is to be generated by performing the following operation:
calculating, based on a sum of the value of the shared processing ranking index corresponding to the target storage unit and a ratio of K to the IO per second parameter of the target storage unit, the processing ranking index of the IO request for which a processing ranking index is to be generated, wherein K is a positive number.
 
3. The storage controller according to claim 2, wherein before calculating the processing ranking index of the IO request for which a processing ranking index is to be generated, said each ranking core is further configured to obtain a current system time; and
said each ranking core calculates the processing ranking index of the IO request for which a processing ranking index is to be generated by performing the following operation:
using, as the processing ranking index of the IO request for which a processing ranking index is to be generated, a larger one of the current system time and the sum of the value of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit.
 
4. The storage controller according to claim 2 or 3, wherein said each ranking core is further configured to:

at a moment after which the processing ranking index is generated for the IO request for which a processing ranking index is to be generated, determine that there is no IO request that has been distributed to said each ranking core, that points to the target storage unit, and for which no index has been generated;

calculate a sum, at the moment, of the value of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit, to use the sum as a pending processing ranking index; and

store the pending processing ranking index in the index queue corresponding to said each ranking core.


 
5. The storage controller according to claim 4, wherein within duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core, an IO request corresponding to a processing ranking index that is greater than the pending processing ranking index and that is comprised in the index queue corresponding to said each ranking core cannot be processed by the request processing core; and
said each ranking core is further configured to remove the pending processing ranking index from the index queue corresponding to said each ranking core if an IO request pointing to the target storage unit is distributed to said each ranking core after the moment or the duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core exceeds a preset threshold.
 
6. The storage controller according to any one of claims 1 to 5, wherein the request processing core periodically processes the IO request corresponding to the smallest processing ranking index in the index queue corresponding to said each ranking core by performing the following operations:

periodically accessing the index queue corresponding to said each ranking core; and

processing an IO request, during each access, corresponding to the smallest processing ranking index in the index queue corresponding to said each ranking core.


 
7. An input/output IO request processing method, wherein the method is performed by a storage controller of a storage system comprising a plurality of storage units, the storage controller comprises a memory device, at least one distribution core, a plurality of ranking cores, and at least one request processing core, the memory device stores a plurality of IO requests, each IO request points to one storage unit, a corresponding shared processing ranking index is further set for each storage unit in the memory device, and the method comprises:

receiving, by each distribution core, an IO request, and distributing the received IO request to the plurality of ranking cores;

obtaining, by each ranking core, an IO request for which a processing ranking index is to be generated and that is distributed by each distribution core;

determining, by said each ranking core, a target storage unit to which the IO request for which a processing ranking index is to be generated points;

obtaining, by said each ranking core, an IO per second parameter of the target storage unit;

generating, by said each ranking core based on the IO per second parameter of the target storage unit and a shared processing ranking index corresponding to the target storage unit, the processing ranking index for the IO request for which a processing ranking index is to be generated;

updating, by said each ranking core, by using the processing ranking index of the IO request for which a processing ranking index is to be generated, the shared processing ranking index corresponding to the target storage unit;

storing, by said each ranking core in an index queue corresponding to said each ranking core, the processing ranking index of the IO request for which a processing ranking index is to be generated, wherein the index queue corresponding to said each ranking core is stored in the memory device and comprises processing ranking indexes that are generated by said each ranking core for the IO requests pointing to the plurality of storage units; and

periodically processing, by each request processing core, an IO request corresponding to a smallest processing ranking index in the index queue corresponding to said each ranking core.


 
8. The method according to claim 7, wherein the generating, by said each ranking core based on the IO per second parameter of the target storage unit and a shared processing ranking index corresponding to the target storage unit, the processing ranking index for the IO request for which a processing ranking index is to be generated comprises:

calculating, based on a sum of the shared processing ranking index corresponding to the target storage unit and a ratio of K to the IO per second parameter of the target storage unit, the processing ranking index of the IO request for which a processing ranking index is to be generated, wherein K is a positive number.


 
9. The method according to claim 8, wherein before the calculating, by said each ranking core, the processing ranking index of the IO request for which a processing ranking index is to be generated, the method further comprises: obtaining, by said each ranking core, a current system time; and
the calculating, by said each ranking core based on a sum of the shared processing ranking index corresponding to the target storage unit and a ratio of K to the IO per second parameter of the target storage unit, the processing ranking index of the IO request for which a processing ranking index is to be generated comprises:
using, by said each ranking core, as the processing ranking index of the IO request for which a processing ranking index is to be generated, a larger one of the current system time and the sum of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit.
 
10. The method according to claim 8 or 9, wherein the method further comprises:

determining, by said each ranking core, at a moment after which the processing ranking index is generated for the IO request for which a processing ranking index is to be generated, that there is no IO request that has been distributed to said each ranking core, that points to the target storage unit, and for which no index has been generated;

calculating, by said each ranking core, a sum, at the moment, of the value of the shared processing ranking index corresponding to the target storage unit and the ratio of K to the IO per second parameter of the target storage unit, to use the sum as a pending processing ranking index; and

storing, by said each ranking core, the pending processing ranking index in the index queue corresponding to said each ranking core.


 
11. The method according to claim 10, wherein within duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core, an IO request corresponding to a processing ranking index that is greater than the pending processing ranking index and that is comprised in the index queue corresponding to said each ranking core cannot be processed by the request processing core; and
said each ranking core removes the pending processing ranking index from the index queue corresponding to said each ranking core if an IO request pointing to the target storage unit is distributed to said each ranking core after the moment or the duration in which the pending processing ranking index exists in the index queue corresponding to said each ranking core exceeds a preset threshold.
 
12. The method according to any one of claims 7 to 11, wherein the periodically processing, by each request processing core, an IO request corresponding to a smallest processing ranking index in the index queue corresponding to said each ranking core comprises:

periodically accessing the index queue corresponding to said each ranking core; and

processing an IO request, during each access, corresponding to the smallest processing ranking index in the index queue corresponding to said each ranking core.


 




Drawing