Storage system and control method for storage system

The storage system optimizes resource allocation by predicting long-term load changes and executing rebalancing processes, ensuring timely addition or removal of controller nodes to reduce power consumption and maintain performance.

JP2024158090APending Publication Date: 2024-11-08HITACHI VANTARA LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
JP2023072959
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-04-27
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing storage systems fail to appropriately allocate resources of the storage controller at the right time, leading to inefficiencies and increased power consumption.

Method used

A storage system with a management server that predicts long-term load changes in controller nodes, calculates the necessary addition or removal of controller nodes based on power performance models, and executes rebalancing processes to optimize resource allocation.

Benefits of technology

This approach allows for timely and efficient allocation of storage controller resources, reducing power consumption and maintaining optimal performance by releasing unnecessary resources and adding necessary ones without delay.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024158090000001_ABST
    Figure 2024158090000001_ABST
Patent Text Reader

Abstract

To allocate a resource of a storage controller appropriately.SOLUTION: A processor of a storage system calculates a long-term load fluctuation forecast that is a forecast of the load fluctuation of multiple controller nodes over a certain period of time in the future based on time-series data of the load of the multiple controller nodes. Then, based on the long-term load fluctuation forecast and load thresholds determined from a power performance model that represents a relation between the load and energy efficiency of the multiple controller nodes, a target time to complete the installation of additional or reduced controller nodes is calculated. Then a rebalance time required for rebalance processing is calculated based on the move amount of data in the rebalance processing that moves data between the multiple drive nodes as they are added or reduced and band information of a path of the data movement. Then, the start time of the rebalance processing from the target time for completion of the addition / reduction and the rebalance time to start the rebalance processing at the start time.SELECTED DRAWING: Figure 5
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a storage system and a control method for a storage system. [Background technology]

[0002] As efforts toward decarbonization are being promoted with the aim of realizing a sustainable society, there is a social demand for energy conservation in IT infrastructure as well. The basis of energy conservation in IT infrastructure is to allocate the necessary and sufficient resources for calculations at the right time. Since resource allocation requires a certain amount of preparation time, it is important to allocate the right amount of resources at the right time.

[0003] Therefore, in order to know the appropriate timing for resource allocation, predictions are sometimes made using historical information. For example, Patent Document 1 discloses a prior art technique for predicting the amount of data to be used in a storage system at a future time, calculating the difference between the predicted amount of data to be used and the currently allocated storage capacity, and increasing or decreasing the storage capacity of the data storage area based on the calculation result. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] International Publication No. 2011 / 092739 Summary of the Invention [Problem to be solved by the invention]

[0005] The storage system includes a data storage area and a storage controller that controls the data storage area. The above-mentioned conventional technology discloses a technology for predicting a required amount of the data storage area and allocating resources to the data storage area based on the prediction result.

[0006] However, the above-mentioned conventional technology only targets resource allocation to data storage areas, not to storage controllers, and therefore cannot allocate an appropriate amount of storage controller resources at the appropriate time.

[0007] The present invention has been made in consideration of the above, and has an object to appropriately allocate resources of a storage controller. [Means for solving the problem]

[0008] As one aspect for solving the above problem, there is provided a storage system having a plurality of drive nodes, a plurality of controller nodes which control I / O of data from a higher-level device to the plurality of drive nodes, and a processor, wherein the processor calculates a long-term load fluctuation prediction, which is a prediction of load fluctuations of the plurality of controller nodes over a certain period of time in the future, based on time series data of the load of the plurality of controller nodes, calculates a target time for completing the expansion or reduction of an operating controller node among the plurality of controller nodes, based on the long-term load fluctuation prediction and a load threshold determined from a power performance model which represents the relationship between the load and energy efficiency of the plurality of controller nodes, calculates a rebalancing time required for the rebalancing process, based on the amount of data movement in a rebalancing process that moves data between the plurality of drive nodes in association with the expansion or reduction, and bandwidth information of the path for the data movement, calculates a start time of the rebalancing process from the target time for expansion or reduction and the rebalancing time, and starts the rebalancing process at the start time. Effect of the Invention

[0009] According to the present invention, power saving can be achieved by appropriately allocating resources of a storage controller. [Brief description of the drawings]

[0010] [Figure 1]FIG. 1 is a diagram showing the configuration of a storage system according to a first embodiment. [Diagram 2] FIG. 2 is a diagram showing the configuration of a controller node according to the first embodiment. [Diagram 3] FIG. 2 is a diagram showing the configuration of a management server according to the first embodiment. [Figure 4A] FIG. 2 is a diagram showing the configuration of a rebalancing planning unit according to the first embodiment and an overview of rebalancing control when a load is reduced. [Figure 4B] FIG. 2 is a diagram showing the configuration of a rebalancing planning unit according to the first embodiment and an overview of rebalancing control when a load increases. [Diagram 5] 6 is a flowchart showing rebalancing and increasing / decreasing processing according to the first embodiment. [Figure 6A] FIG. 4 is a diagram showing a power performance model (when the load is reduced) according to the first embodiment. [Figure 6B] FIG. 4 is a diagram showing a power performance model (when the load increases) according to the first embodiment. [Figure 7] FIG. 4 is a diagram illustrating a rebalancing operation according to the first embodiment. [Figure 8] FIG. 11 is a diagram showing the configuration of a storage system according to a second embodiment. [Figure 9] FIG. 11 is a diagram showing the configuration of a storage node according to the second embodiment. [Figure 10] FIG. 11 is a diagram illustrating a rebalancing operation according to the second embodiment. [Figure 11] FIG. 11 is a diagram showing the configuration of a storage system including a plurality of storage systems according to a third embodiment. [Figure 12] FIG. 13 is an explanatory diagram of load control for suppressing the system load including rebalancing to a system load allowable value or less according to the fourth embodiment. [Figure 13] FIG. 13 is an explanatory diagram of load control in which a short-term load fluctuation range is excluded from a long-term load fluctuation prediction according to the fifth embodiment. [Figure 14] Computer hardware diagram. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In this embodiment, the same components are generally designated by the same reference numerals, and repeated explanations will be omitted. Note that this embodiment is merely an example for realizing the present invention, and does not limit the technical scope of the present invention.

[0012] In the following explanation, the processing performed by a program may be described. A computer performs processing defined by a program using a processor (e.g., a CPU (Central Processing Unit), a GPU (Graphics Processing Unit)) while using memory in a main storage device, etc. Therefore, the processor may be the entity that performs processing by executing a program. The processor executes a program to realize a functional unit that performs processing.

[0013] Similarly, the subject of the processing performed by executing the program may be a controller, device, system, computer, or node having a processor. The subject of the processing performed by executing the program may be a computing unit, and may include a dedicated circuit that performs specific processing. The dedicated circuit is, for example, an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit).

[0014] In the following description, a program may be installed in a computer from a program source. The program source may be, for example, a program distribution server or a non-transitory storage medium readable by a computer. When the program source is a program distribution server, the program distribution server may include a processor and a storage resource (storage) for storing a program to be distributed, and the processor of the program distribution server may distribute the program to be distributed to other computers. In addition, in an embodiment, two or more programs may be realized as one program, and one program may be realized as two or more programs.

[0015] [Embodiment 1] (Configuration of Storage System 1S According to First Embodiment) 1 is a diagram showing the configuration of a storage system 1S according to embodiment 1. The storage system 1S is a separated type storage system in which a storage controller (controller node) and a drive node are separated.

[0016] The storage system 1S includes a plurality of compute nodes 400, a plurality of controller nodes 100, and a plurality of drive nodes 200. In the storage system 1S, these are connected via a network switch 300 between the controller nodes and a network switch 320 between the compute nodes.

[0017] An access request issued by an application program executed on the compute node 400 is processed for data in the drive node 200 under the control of the controller node 100. For system management, a management server 410 and a system environment external server 420 are connected via a management network switch 330. To ensure system availability, all nodes and paths are made redundant, although some are not shown in the figure.

[0018] (Configuration of the controller node 100 according to the first embodiment) FIG. 2 is a diagram illustrating a configuration of the controller node 100 according to the first embodiment.

[0019] The controller node 100 has a processor 110, a memory 115, a management port 120, a front-end port 130, and a back-end port 140. The controller node 100 provides functions such as data input / output processing, data encryption, data compression, snapshot creation, and data virtualization in accordance with instructions from the compute node 400. The controller node 100 is connected to the compute node 400 via the front-end port 130, to the drive node 200 via the back-end port 140, and to the management server 410 via the management port 120.

[0020] The processor 110 inputs and outputs data via the front-end port switch 160 and the front-end port 130. The processor 110 also inputs and outputs data via the back-end port switch 170 and the back-end port 140.

[0021] (Configuration of the Management Server 410 According to the First Embodiment) FIG. 3 is a diagram showing the configuration of the management server 410 according to the first embodiment.

[0022] The management server 410 is connected to the storage system 1S via a management network switch 330 and a management network (not shown). In Fig. 3, only the functional configuration of the management server 410 related to this embodiment is shown and described, and the illustration and description of other functional configurations are omitted.

[0023] The management server 410 is configured to include a long-term load fluctuation prediction unit 411, a rebalancing planning unit 412, a rebalancing implementation unit 413, an expansion / reduction implementation unit 414, a communication unit 415, and a power performance model 20. The management server 410 also has a display unit that displays information to an operator, and an input unit such as a keyboard that accepts information input by the operator. The power performance model 20 is stored in a predetermined storage area. In the power performance model 20, the operator can edit controller expansion thresholds 3A (system loads L1 to L3) and controller expansion thresholds 3B (system loads L4 to L6), which will be described later, via the display unit and the input unit.

[0024] The communication unit 415 is a communication interface that enables the management server 410 to communicate with the controller node 100 , the drive node 200 , the compute node 400 , and the storage node 500 via the management network switch 330 .

[0025] The long-term load fluctuation prediction unit 411 predicts the trend of the amount of fluctuation when observing fluctuations in the load on the controller node 100 (hereinafter referred to as "controller load") over a certain span of time (for example, several weeks). When a state in which the trend can be considered constant continues for a certain period of time, the long-term load fluctuation prediction unit 411 determines that there is a long-term load fluctuation in the controller load that represents fluctuations in the load from a long-term perspective. In this way, the long-term load fluctuation prediction unit 411 predicts whether the controller load viewed over a certain long span is on an increasing trend or a decreasing trend.

[0026] 5 may be executed by a plurality of processors 110 of each controller node 100 through distributed cooperative processing instead of the management server 410. That is, the rebalancing and expansion / reduction processing is executed by the processor 1001 of the management server 410 or the processor 110 of the controller node 100.

[0027] (Configuration of rebalancing planning unit 412) First, rebalancing control when the load is reduced will be described. Fig. 4A is a diagram showing the configuration of the rebalancing plan unit 412 according to the first embodiment and an overview of rebalancing control when the load is reduced.

[0028] The rebalancing planning unit 412 has a rebalancing time prediction unit 11 and a rebalancing start time prediction unit 12. The rebalancing planning unit 412 also stores each piece of data, such as constraint conditions 13, data movement amount 14, and bandwidth information 15, in a storage area.

[0029] The controller node 100 constantly measures and accumulates the processing load (controller load) of its own processor 110 as load time series data 1. The rebalancing plan unit 412 calculates a long-term load fluctuation prediction 2 by analyzing the load time series data 1.

[0030] The controller load fluctuates according to the amount of work based on requests from higher-level devices such as hosts to the storage system 1S. The long-term load fluctuation prediction 2 is a prediction of the fluctuation trend of the controller load that is predicted to continue for a certain period of time in the future (for example, several weeks).

[0031] There is a limit to the amount of controller load fluctuation that can be covered by one controller node 100. For this reason, when the long-term load fluctuation prediction 2 exceeds the controller increase / decrease threshold (controller increase / decrease threshold 3A or controller increase / decrease threshold 3B), it is necessary to increase or decrease the number of operating controllers. In order to release unnecessary resources without delay and maintain a necessary and sufficient resource allocation, the goal is to complete the increase / decrease of controllers at the timing when the long-term load fluctuation prediction 2 exceeds the controller increase / decrease threshold.

[0032] The controller increase / decrease threshold becomes controller decrease threshold 3A when the load time series data 1 decreases as shown in the graph of Fig. 4A, and becomes controller increase threshold 3B when the load time series data 1 increases as shown in the graph of Fig. 4B. The controller decrease threshold 3A and the controller increase threshold 3B are generally different.

[0033] The rebalancing planning unit 412 judges whether the long-term load fluctuation prediction 2 exceeds the controller add / drop threshold (controller add / drop threshold 3A or controller add / drop threshold 3B). The rebalancing planning unit 412 then determines the timing at which the long-term load fluctuation prediction 2 exceeds the controller add / drop threshold 3A as the target time (target add / drop completion time t12) for completing the add / drop of the controller node 100. The target add / drop completion time t12 can be calculated from the intersection of the long-term load fluctuation prediction 2 and the controller add / drop threshold 3A.

[0034] On the other hand, in order to remove or add a controller node 100, it is necessary to migrate the data under the jurisdiction of the controller node 100 to the jurisdiction of another controller node 100. This is called controller rebalancing. Rebalancing generally includes the movement of actual data and the movement of metadata. Actual data is the actual data stored in the storage area of ​​the storage system. Metadata is data used to manage the actual data. In this embodiment, the drive nodes 200 are not added or removed, so the movement of actual data does not occur, and only the movement of metadata occurs.

[0035] The rebalancing time 4A required for rebalancing the controller node 100 depends on the constraint conditions 13, the amount of data movement 14 during rebalancing, and the bandwidth information 15 of the path used for this data movement. The constraint conditions 13 include, for example, time constraints such as the time periods during which data can be moved, and physical constraints such as the order in which data is moved. The rebalancing time 4A may also be influenced by information other than the constraint conditions 13, the amount of data movement 14, and the bandwidth information 15.

[0036] The rebalancing time prediction unit 11 predicts the rebalancing time 4A based on the constraint conditions 13, the data movement amount 14, and the bandwidth information 15. The rebalancing start time prediction unit 12 sets the rebalancing start time t11 to be before the time calculated by subtracting the rebalancing time 4A from the target reduction completion time t12. In this way, the rebalancing plan unit 412 creates a rebalancing plan including the rebalancing start time t11, the rebalancing time 4A, and the target reduction completion time t12.

[0037] Next, rebalancing control when the load increases will be described. Fig. 4B is a diagram showing the configuration of the rebalancing planning unit 412 according to the first embodiment and an overview of rebalancing control when the load increases.

[0038] The timing when the long-term load fluctuation prediction 2 exceeds the controller addition threshold 3B is set as the target time (target addition completion time t22) for completing the addition of the controller node 100. The target addition completion time t22 can be calculated from the intersection of the long-term load fluctuation prediction 2 and the controller addition threshold 3B.

[0039] A rebalancing time 4B required for rebalancing of the controller node 100 depends on constraint conditions 13, data movement amount 14 during rebalancing, and bandwidth information 15 used for this data movement. The rebalancing time 4B may also be influenced by information other than constraint conditions 13, data movement amount 14, and bandwidth information 15.

[0040] The rebalancing time prediction unit 11 predicts the rebalancing time 4B based on the constraint conditions 13, the data movement amount 14, and the bandwidth information 15. The rebalancing start time prediction unit 12 sets the rebalancing start time t21 to be before the time calculated by subtracting the rebalancing time 4B from the target addition completion time t22.

[0041] Returning to the explanation of Fig. 3, the rebalancing implementation unit 413 transmits an instruction to the controller node 100 to implement rebalancing in accordance with the rebalancing plan created by the rebalancing plan unit 412. When rebalancing is completed, the rebalancing implementation unit 413 receives a rebalancing completion notification from the controller node 100.

[0042] When the rebalancing implementation unit 413 receives a completion notification of rebalancing, the expansion / reduction implementation unit 414 transmits an instruction to each controller node 100 to expand or reduce the number of controller nodes 100. When the expansion / reduction of the controller nodes 100 is completed, the expansion / reduction implementation unit 414 receives a completion notification of the expansion / reduction from the controller node 100.

[0043] (Rebalancing and expansion / reduction processing according to the embodiment) 5 is a flowchart showing the rebalancing and expansion / reduction processing according to embodiment 1. The rebalancing and expansion / reduction processing according to the embodiment is periodically executed by the management server 410.

[0044] First, in step S11, the long-term load fluctuation prediction unit 411 calculates the long-term load fluctuation prediction 2 based on the load time series data 1. Next, in step S12, the long-term load fluctuation prediction unit 411 judges whether to increase or decrease the number of controller nodes 100. Specifically, the long-term load fluctuation prediction unit 411 judges whether the trend of the long-term load fluctuation prediction 2 of the controller is expected to exceed the controller increase / decrease threshold within a certain time. If the trend curve of the long-term load fluctuation prediction 2 is expected to exceed the controller increase / decrease threshold within a certain time (step S12 YES), the long-term load fluctuation prediction unit 411 judges that an increase / decrease is necessary and moves the process to step S13. On the other hand, if the trend curve of the long-term load fluctuation prediction 2 is not expected to exceed the controller increase / decrease threshold within a certain time (step S12 NO), the long-term load fluctuation prediction unit 411 judges that an increase / decrease is unnecessary and ends this rebalancing and increase / decrease processing.

[0045] In step S13, the rebalancing planning unit 412 creates a rebalancing plan based on the latest information of the long-term load fluctuation prediction 2, in which it has been determined in step S13 that the trend curve will exceed the controller increase / decrease threshold within a certain period of time, and on the power performance model 20. In creating the plan, the controller increase / decrease threshold required is given by the power performance model 20. Details of the power performance model 20 will be described later with reference to Figs. 6A and 6B.

[0046] Next, in step S14, the rebalancing plan unit 412 determines whether to continue rebalancing. Since rebalancing requires a certain amount of time or more, it is necessary to periodically check whether there is no change in the trend of decrease or increase in the long-term load fluctuation prediction 2 and whether the rebalancing plan created in step S13 may be implemented as planned. In addition, for example, it may be checked whether the cost of increasing or decreasing the controller node 100 (electricity cost for data movement for rebalancing, etc.) is worth the operating cost for a predetermined period after the controller node 100 is increased or decreased. For example, if the cost of decreasing the controller node 100 is greater than the reduction in the operating cost of the controller node 100 after the decrease, the cost is not worth it, so the decrease in the controller node 100 should not be implemented. In addition, if the controller load does not increase more than expected even if the controller node 100 is increased, the cost is not worth it, so the increase in the controller node 100 should not be implemented. In this way, based on the cost comparison result, it is determined whether to decrease or increase the controller node 100.

[0047] If the rebalancing is to be continued (step S14: YES), the rebalancing planning unit 412 proceeds to step S15, and if the rebalancing is to be interrupted (step S14: NO), the rebalancing planning unit 412 ends this rebalancing and expansion / reduction processing.

[0048] In step S15, the rebalancing execution unit 413 transmits an instruction to the controller node 100 to execute the movement of actual data and / or metadata as the rebalancing process.

[0049] Next, in step S16, the rebalancing implementation unit 413 determines the completion of rebalancing by determining whether or not a rebalancing completion notification has been received from the controller node 100. Step S16 is executed when a predetermined time has elapsed after execution of step S15.

[0050] If rebalancing is complete (step S16 YES), the rebalancing implementation unit 413 proceeds to step S17, and if rebalancing is not complete (step S16 NO), the rebalancing implementation unit 413 proceeds to step S13. In step S13 to which the process is transferred from step S16, the rebalancing plan unit 412 modifies the rebalancing plan based on the latest information of the long-term load fluctuation prediction 2 and the power performance model 20, and executes steps S14 to S16 based on the modified rebalancing plan. By modifying the rebalancing plan in the process flow from step S16 to step S13, it is possible to avoid a situation in which the rebalancing plan created in step S13 and the current state of the long-term load fluctuation prediction 2 diverge significantly.

[0051] In step S17, the expansion / reduction implementation unit 414 transmits an instruction to the controller node 100 to expand or reduce the controller node 100. When the expansion / reduction of the controller node 100 is completed, the expansion / reduction implementation unit 414 receives a completion notification of the expansion / reduction from the controller node 100.

[0052] As described above, by starting the rebalancing required for adding or removing controllers in accordance with the target time for completing the controller adding or removing, it is possible to add or remove controllers at the required timing without delay.

[0053] (Power performance model 20 according to the embodiment) Next, the power performance model 20 according to the embodiment will be described. Fig. 6A is a diagram showing the power performance model 20 (when load decreases) according to the first embodiment. Fig. 6B is a diagram showing the power performance model 20 (when load increases) according to the first embodiment. In Figs. 6A and 6B, the horizontal axis represents the scale of the system load (controller load), and the vertical axis represents the scale of the power or energy efficiency. In Figs. 6A and 6B, the bar graphs represent the power at each load, and the line graphs represent the energy efficiency at each load.

[0054] 6A and 6B, the number of controller nodes 100 can be varied between 1 and 4, and three thresholds are provided for when the controller load is decreasing and when the load is increasing. The thresholds are determined according to the number of controller nodes 100 that are variable.

[0055] In the power performance model 20, the controller removal threshold 3A and the controller addition threshold 3B are adjusted according to the controller load of the controller node 100. This provides a function for optimizing the trade-off between the performance and power efficiency of the controller node 100 in accordance with system requirements.

[0056] As shown in Fig. 6A, in the power performance model 20 when the load of the controller is reduced, the controller reduction threshold 3A is set so that the number of controllers is reduced at relatively wide intervals such as system loads L1, L2, and L3. The system power decreases in accordance with the reduction in the number of controllers. In this way, the performance of the controller is reduced relatively gradually in response to the reduction in the load of the controller.

[0057] When the controller load is reduced, the controller performance is scaled down earlier than when the load is increased. This is because a high controller load is more energy efficient but has poorer response performance.

[0058] On the other hand, as shown in Fig. 6B, in the power performance model 20 when the controller load increases, the controller addition threshold 3B is set so that the number of controllers increases at narrower intervals such as system loads L4, L5, and L6 compared to system loads L1, L2, and L3. The system power increases as the number of controllers increases. In this way, the controller performance is expanded more quickly in response to an increase in the controller load than when the load decreases.

[0059] In this way, by adjusting the controller reduction threshold 3A and the controller expansion threshold 3B, it is possible to adjust the trade-off between power saving and performance. In other words, it is possible to provide the necessary performance without delay when the load is increasing, and to release unnecessary resources without delay when the load is decreasing, thereby realizing power saving. The controller reduction threshold 3A and the controller expansion threshold 3B may be adjusted according to the power composition ratio. When the proportion of renewable energy is high, the amount of CO2 emissions is small, so that the controller reduction threshold 3A and the controller expansion threshold 3B can be set low to make expansion easier and reduction more difficult, and to provide a margin of performance. The controller expansion / reduction threshold of the power performance model 20 may be made changeable at the time of system design or operation according to the business requirements of the upper device.

[0060] (Rebalancing according to embodiment 1) FIG. 7 is a diagram illustrating the rebalancing operation according to the first embodiment.

[0061] The history of input / output processed by the controller node 100 is collected in the management server 410 as load time series data 1. The management server 410 calculates a long-term load fluctuation prediction 2 from the collected load time series data 1 and plans predictive rebalancing (step S13 (FIG. 5)). The result is sent to each controller node 100, and predictive rebalancing is executed (step S15 (FIG. 5)).

[0062] 7, the drive node 200, which is a data storage area, is separated from the controller node 100 and connected by a network switch 300 between controller nodes. Therefore, when rebalancing is performed due to a change in the controller node 100 that manages the drive node 200, it is not necessary to move the actual data, but it is necessary to move the metadata, which is management information for the actual data. After the necessary metadata is moved, the access path to the controller node 100 is switched, and the expansion or reduction of the controller node is completed.

[0063] As described above, by completing the metadata migration in advance based on the long-term load fluctuation prediction 2, it becomes possible to add or remove controllers at the required timing without delay.

[0064] (Effects of the First Embodiment) In the above-mentioned first embodiment, overhead processing required for adding or removing storage controllers, such as data movement of metadata, is performed in advance based on a long-term load fluctuation prediction. Therefore, controllers can be added or removed without delay at the required timing, and auto-scaling of the storage system can be performed. In addition, since the storage controllers can be added or removed without delay in response to load fluctuations, necessary and sufficient resources can be continuously provided, and unnecessary resources can be quickly released, realizing power saving.

[0065] [Embodiment 2] 8 is a diagram showing the configuration of a storage system 2S according to embodiment 2. The storage system 2S includes a storage node in which a storage controller and a drive node are integrated. In embodiment 2, the process of adding or removing a storage node in the storage system 2S is the same as the process in embodiment 1 in which the controller node is replaced with a storage node, and only the differences will be described below.

[0066] In the storage system 2S, a plurality of compute nodes 400 and a plurality of storage nodes 500 are connected via an inter-storage node network switch 310 and an inter-compute node network switch 320. For system management, a management server 410 and a system environment external server 420 are connected via a management network switch 330. To ensure system availability, all nodes and paths are made redundant, although some are omitted from the illustration.

[0067] (Configuration of the storage node 500 according to the second embodiment) 9 is a diagram showing the configuration of a storage node 500 according to embodiment 2. The storage node 500 is a device in which a controller node 100 and a drive node 210 are integrated together.

[0068] (Rebalancing according to the second embodiment) FIG. 10 is a diagram illustrating the rebalancing operation according to the second embodiment.

[0069] The history of input / output processed by the controller node 100 in the storage node 500 is collected by the management server 410 as load time series data 1. The management server 410 calculates a long-term load fluctuation prediction 2 from the collected load time series data 1 and plans predictive rebalancing (step S13 (FIG. 5)). The result is sent to the controller node 100, and predictive rebalancing is executed (step S15 (FIG. 5)).

[0070] 10, the drive node 210, which is a data storage area, and the controller node 100 are integrated within the storage node 500. Therefore, the actual data needs to be moved during rebalancing due to the addition or removal of a controller. After the necessary actual data (and associated metadata) have been moved, the access path to the storage node 500 is switched, and the addition or removal of the storage node 500 is completed.

[0071] 10, the transitions between the states of one controller configuration, two controller configuration, three controller configuration, and four controller configuration are performed in units of storage nodes 500. This is because the drive nodes 210 and the controller nodes 100 are integrated within the storage node 500. Therefore, when a storage node 500 is added, new data is written to the added storage node 500. On the other hand, when a storage node 500 is removed, the actual data and metadata stored in the storage node 500 to be removed are moved to other storage nodes 500 that are not to be removed.

[0072] (Effects of the second embodiment) In the above-mentioned embodiment 2, even in a storage node in which a controller node and a drive node are integrated, the movement of actual data and metadata is completed in advance based on the long-term load fluctuation prediction 2. Therefore, the storage controller can be added or removed without delay at the required timing, and power saving can be realized.

[0073] [Embodiment 3] Fig. 11 is a diagram showing the configuration of a storage system 3S including a plurality of storage systems 1S and / or 2S according to the third embodiment. In Fig. 11, the storage system 3S is illustrated as including one each of storage systems 1S and 2S, but the number of these can be changed. The storage system 3S may be configured to include only a plurality of storage systems 1S, may be configured to include only a plurality of storage systems 2S, or may be configured to include a plurality of storage systems 1S and 2S.

[0074] The storage system 3S is a hybrid storage system configured by connecting a plurality of storage clusters of the storage systems 1S and / or 2S via an inter-cluster switch 340. The storage system 3S is used for the purpose of optimizing data placement by using the storage systems 1S and 2S differently according to the data usage and data lifecycle.

[0075] For example, a use case may be considered in which data is moved from a storage cluster with high data storage costs to a storage cluster with low data storage costs as the data life cycle progresses. In this case, the management server 410 performs predictive reduction according to embodiment 1 or 2 in the source storage cluster, and performs predictive expansion according to embodiment 1 or 2 in the destination storage cluster. Furthermore, the management server 410 performs predictive expansion according to embodiment 1 or 2 between the source storage cluster and the destination storage cluster. This makes it possible to allocate and release necessary and sufficient resources without delay in the storage system 3S.

[0076] In each storage cluster, the workload associated with the operation of the project causes load fluctuations in the storage controllers related to data. Storage controllers are expanded or reduced based on long-term load fluctuation predictions related to these load fluctuations. Whether data migration associated with rebalancing is performed within the own storage cluster or to another storage cluster differs depending on the use case of the project and workload.

[0077] (Effects of the Third Embodiment) In the above-described third embodiment, in the hybrid storage system, the storage controllers can be added or removed in units of each storage cluster, and the storage controllers can be added or removed across multiple storage clusters.

[0078] [Embodiment 4] In the above-described first to third embodiments, the controller nodes 100 are increased or decreased, including rebalancing, depending only on the timing of increase or decrease in the controller load (system load), without taking into consideration an increase in the rebalancing processing load on the controller node 100.

[0079] However, if the load on the controller increases further due to rebalancing, problems such as a decrease in I / O performance for higher-level systems such as hosts can occur.

[0080] Therefore, in the fourth embodiment, a limit is set for the implementation period of the rebalancing in the first to third embodiments, and the overall system load including the rebalancing is controlled so as not to exceed a threshold.

[0081] (Load Control According to Fourth Embodiment) Fig. 12 is an explanatory diagram of load control for suppressing the system load including rebalancing to a system load tolerance value or less according to the fourth embodiment. Fig. 12 shows a short-term load fluctuation range 7 of the load time-series data 1. Here, "short-term" refers to, for example, about 2 to 3 days.

[0082] In the fourth embodiment, when rebalancing is performed, specific constraint conditions are set to optimize the system operation. In order to mitigate the performance degradation effect of normal I / O etc. due to rebalancing, the controller node 100 temporarily stops the rebalancing process when the load fluctuation of the controller node 100 exceeds the system load tolerance value 6.

[0083] (Effects of the fourth embodiment) In the above-mentioned fourth embodiment, when the load fluctuation of the above-mentioned controller node 100 exceeds the system load tolerance 6, it is possible to balance the rebalancing process and normal processing by controlling to temporarily stop the rebalancing process. In addition, performing this control increases the time until the rebalancing is completed, but by calculating the rebalancing time taking into account the increase in time, it becomes possible to execute the rebalancing in accordance with the target time for expansion / reduction, thereby realizing power saving.

[0084] [Embodiment 5] In the above-mentioned embodiments 1 to 3, the short-term load fluctuation of the long-term load fluctuation prediction 2 of the controller node 100 is not taken into consideration, and the timing when the long-term load fluctuation prediction 2 reaches, for example, a controller reduction threshold value 3A is set as the reduction completion target time t12 as shown in FIG. 13. Here, "short term" is, for example, about 2 to 3 days. However, if the long-term load fluctuation prediction 2 fluctuates significantly in the short term, it is desirable to set the reduction target time to a timing with a margin that allows for this fluctuation. This problem also occurs when calculating the expansion completion target time t22.

[0085] In the fifth embodiment, when rebalancing is performed, specific constraint conditions are set to optimize the system operation. That is, in the fifth embodiment, when calculating the target time for increasing or decreasing the capacity, the short-term load fluctuation range 7 of the long-term load fluctuation prediction 2 is taken into consideration, and the target time for increasing or decreasing the capacity is controlled to be calculated at the maximum value of the short-term load fluctuation range 7.

[0086] (Load Control According to the Fifth Embodiment) Fig. 13 is an explanatory diagram of load control that excludes a short-term load fluctuation range 7 from a long-term load fluctuation prediction 2 according to the fifth embodiment. Fig. 13 shows a case where the controller load decreases. Usually, the intersection of the long-term load fluctuation prediction 2 and the controller reduction threshold value 3A is set as the reduction completion target time t12.

[0087] However, when the short-term load fluctuation range 7 is large, it is possible that the load time series data 1 will fluctuate above and below the controller reduction threshold 3A in the short term, making it difficult to plan a stable rebalancing. Therefore, in order to improve the prediction accuracy of the reduction completion time, the short-term load fluctuation range 7 is calculated, and the intersection point between the straight line representing the maximum value of the short-term load fluctuation range 7 and the controller reduction threshold 3A is set as the target reduction completion time t13.

[0088] (Effects of the fifth embodiment) In the above-mentioned embodiment 5, by executing the reduction with the target reduction completion time t13 as the target, rebalancing and the reduction of the controller node 100 can be stably performed at the target reduction completion time t13 with a safety margin that takes into account the short-term load fluctuation range 7.

[0089] Although FIG. 13 shows a case where the controller node 100 is removed, the addition completion time when the controller node 100 is added can be calculated in the same manner.

[0090] (Computer 1000 Hardware) 14 is a hardware diagram showing the configuration of a computer 1000. For example, the compute node 400, the management server 410, the system environment external server 420, and the storage node 500 are realized by the computer 1000.

[0091] The computer 1000 includes a processor 1001 including a CPU, a main memory device 1002, an auxiliary memory device 1003, a network interface 1004, an input device 1005, and an output device 1006, all of which are interconnected via an internal communication line 1009 such as a bus.

[0092] The processor 1001 controls the overall operation of the computer 1000. The main memory device 1002 is composed of, for example, a volatile semiconductor memory, and is used as a work memory for the processor 1001. The auxiliary memory device 1003 is an example of a non-transitory storage medium, and is composed of a large-capacity non-volatile storage device such as a hard disk device, a solid state drive (SSD), or a flash memory, and is used to store various programs and data for a long period of time.

[0093] An executable program 1100 stored in the auxiliary storage device 1003 is loaded into the main storage device 1002 when the computer 1000 is started up or when necessary, and the executable program 1100 loaded into the main storage device 1002 is executed by the processor 1001. This realizes a system that executes various processes and each functional unit (long-term load fluctuation prediction unit 411, rebalancing planning unit 412, rebalancing implementation unit 413, and expansion / reduction implementation unit 414).

[0094] The executable program 1100 may be recorded on a non-transitory recording medium, read from the non-transitory recording medium by a media reading device, and loaded into the primary storage device 1002. Alternatively, the executable program 1100 may be obtained from an external computer via a network, and loaded into the primary storage device 1002.

[0095] The network interface 1004 is an interface device for connecting the computer 1000 to each network in the system or for communicating with other computers. The network interface 1004 is formed of, for example, a network interface card (NIC) for a wired local area network (LAN) or a wireless LAN.

[0096] The input device 1005 is made up of a keyboard, a pointing device such as a mouse, and the like, and is used by the user to input various instructions and information to the computer 1000 .

[0097] The output device 1006 is composed of a display device such as a liquid crystal display or an organic EL (Electro Luminescence) display, and an audio output device such as a speaker, and is used to present necessary information to the user when necessary.

[0098] Although the embodiments of the present disclosure have been described above in detail, the present disclosure is not limited to the above-described embodiments, and various modifications can be made without departing from the spirit of the present disclosure. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present disclosure is not necessarily limited to those having all of the configurations described. In addition, it is possible to add, delete, or replace part of the configuration of the above-described embodiments with other configurations.

[0099] Furthermore, the above-mentioned configurations, functional units, processing units, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. The above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the program, table, file, etc. that realizes each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), an IC card, an SD card, a DVD, or other recording medium.

[0100] In addition, in each of the above figures, the control lines and information lines are shown as those considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the actual implementation. For example, it may be considered that almost all the components are actually connected to each other.

[0101] The above-mentioned arrangement of the functions and data of the compute node 400, the management server 410, the system environment external server 420, and the storage node 500 is merely an example. The arrangement of the functions and data may be changed to an optimal arrangement in terms of the performance of the hardware and software, processing efficiency, communication efficiency, and the like. [Explanation of symbols]

[0102] 100: controller node, 110: processor, 200: drive node, 210: drive, 400: compute node, 410: management server, 411: long-term load fluctuation prediction unit, 412: rebalancing planning unit, 413: rebalancing implementation unit, 414: expansion / reduction implementation unit, 500: storage node, 1001: processor, 1002: main memory device.

Claims

1. A storage system having a plurality of drive nodes, a plurality of controller nodes that control data I / O from a host device to the plurality of drive nodes, and a processor, The processor: Calculating a long-term load fluctuation prediction, which is a prediction of a fluctuation in the load of the plurality of controller nodes over a certain period of time in the future, based on the time-series data of the load of the plurality of controller nodes; calculating a target time for completing the expansion or reduction of an operating controller node among the plurality of controller nodes based on the long-term load fluctuation prediction and a load threshold determined from a power performance model that represents the relationship between the load and energy efficiency of the plurality of controller nodes; calculating a rebalancing time required for the rebalancing process based on the amount of data moved in the rebalancing process that moves data between the plurality of drive nodes in association with the addition or removal and bandwidth information of the path for the data movement; calculating a start time for the rebalancing process from the target time for completion of the expansion / reduction and the rebalancing time; The rebalancing process starts at the start time. A storage system comprising:

2. 2. The storage system according to claim 1, The processor: When starting the rebalancing process, a cost required for the data movement in the rebalancing process is compared with a cost required for operating the plurality of controller nodes for a predetermined period when the expansion or reduction is implemented; The rebalancing process is initiated based on the cost comparison result. A storage system comprising:

3. 2. The storage system according to claim 1, The processor: determining whether the rebalancing process is complete; When it is determined that the rebalancing process has been completed, the controller node is expanded or expanded, and If it is determined that the rebalancing process has not been completed, the long-term load fluctuation prediction is corrected, and the rebalancing process is performed based on the corrected long-term load fluctuation prediction. A storage system comprising:

4. 2. The storage system according to claim 1, the drive node and the controller node are independent nodes, The data to be moved between the plurality of drive nodes in association with the expansion or reduction is metadata used for managing actual data. A storage system comprising:

5. 2. The storage system according to claim 1, the drive node and the controller node are integrated to form a storage node, The data to be moved between the plurality of drive nodes in association with the expansion or reduction includes actual data and metadata used to manage the actual data. A storage system comprising:

6. 2. The storage system according to claim 1, The processor: The rebalancing process is performed only during a period in which the load fluctuating in a predetermined short period is equal to or less than an allowable value in the long-term load fluctuation prediction. A storage system comprising:

7. 2. The storage system according to claim 1, The processor: The target time for completing the expansion or reduction is calculated based on the long-term load fluctuation prediction and a load that fluctuates in a predetermined short period in the long-term load fluctuation prediction. A storage system comprising:

8. 2. The storage system according to claim 1, The load threshold is configured to be externally adjustable depending on whether power saving or performance of the plurality of controller nodes is to be prioritized. A storage system comprising:

9. The storage system according to any one of claims 1 to 8, an inter-cluster switch that connects the plurality of storage systems; The rebalancing process and the expansion or reduction are carried out within each storage system of the plurality of storage systems, and the rebalancing process and the expansion or reduction are carried out between the plurality of storage systems. A storage system comprising:

10. A storage system control method executed by a storage system having a plurality of drive nodes, a plurality of controller nodes that control data I / O from a host device to the plurality of drive nodes, and a processor, comprising: the processor: Calculating a long-term load fluctuation prediction, which is a prediction of a fluctuation in the load of the plurality of controller nodes over a certain period of time in the future, based on the time-series data of the load of the plurality of controller nodes; calculating a target time for completing the expansion or reduction of an operating controller node among the plurality of controller nodes based on the long-term load fluctuation prediction and a load threshold determined from a power performance model that represents the relationship between the load and energy efficiency of the plurality of controller nodes; calculating a rebalancing time required for the rebalancing process based on the amount of data moved in the rebalancing process that moves data between the plurality of drive nodes in association with the addition or removal and bandwidth information of the path for the data movement; calculating a start time for the rebalancing process from the target time for completion of the expansion / reduction and the rebalancing time; The rebalancing process starts at the start time. A control method for a storage system characterized by including each process.

Citation Information

Patent Citations

  • Data arrangement control program, data arrangement controller, data arrangement control method, and multi-node storage system

    JP2009157571A

  • Storage system, storage device, method for optimizing the storage area of ​​a storage system

    JP2012533099A

  • Distributed storage system and management method

    JP2022150953A

  • Distributed storage system and volume management method

    JP2023016223A

  • Geological Allocation of Storage Space

    US20200133532A1