SSD performance improvements in each power state through deterministic utilization control

By employing computational algorithms for static resource allocation in SSDs, the problem of unpredictable power and performance characteristics under different capacities is solved, achieving efficient power consumption management in NVMe SSDs and improving performance and stability per watt.

CN117667382BActive Publication Date: 2026-11-17SK HYNIX INC
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202211583127.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2022-09-08
Filing Date
2022-12-09
Publication Date
2026-11-17
Estimated Expiration
2042-12-09

AI Technical Summary

Technical Problem

Existing SSDs lack the ability to predict power and performance characteristics at different capacities, resulting in unpredictable power consumption and unstable performance, especially under high load peak conditions, making it difficult to achieve optimized power management.

Method used

A computational algorithm is employed to dynamically control the allocation and release of hardware resources through a static resource allocation table. This allows for precise control based on different workloads and power states, predicting performance per watt, and is applicable to NVMe SSDs.

Benefits of technology

It achieves predictable SSD performance under different capacities and load conditions, optimizes throughput and power consumption per watt, and improves the energy efficiency management capabilities of SSDs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117667382B_ABST
    Figure CN117667382B_ABST
Patent Text Reader

Abstract

A method for operating a system including a host and at least one solid-state drive (SSD). The method identifies workloads associated with the SSD, identifies the power state of the SSD, and controls the allocation and / or unallocation of hardware resources for the identified workloads based on budget targets for the power state.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of this disclosure relate to a scheme for controlling power utilization in a memory system. Background Technology

[0002] The computing environment paradigm has shifted to ubiquitous computing systems that can be used anytime, anywhere. Consequently, the use of portable electronic devices such as mobile phones, digital cameras, and laptops has increased rapidly. These portable electronic devices typically use memory systems with memory devices, also known as data storage devices. The data storage device serves as either the main memory or auxiliary memory in the portable electronic device.

[0003] Memory systems using memory devices offer superior stability, durability, high data access speeds, and low power consumption due to the absence of moving parts. Examples of memory systems with these advantages include Universal Serial Bus (USB) memory devices, memory cards with various interfaces such as Universal Flash Memory (UFS), and solid-state drives (SSDs). The memory system receives commands associated with the memory device from the host and processes these commands on the memory device. Summary of the Invention

[0004] Various aspects of the present invention include a system and method for operating a data storage system.

[0005] On one hand, a method is provided for operating a system including a host and at least one solid-state drive (SSD), the method comprising: identifying a workload associated with the SSD; identifying a power state of the SSD in response to an operation command received from the host; and controlling the allocation and / or deallocation of hardware resources for the identified workload based on a budgeted target of the identified power state.

[0006] On the other hand, a system for operating a data processing system is provided, the system comprising: a host; a memory storage device; and a controller included in the memory storage device, the controller being configured to: identify a workload associated with an SSD, identify the power state of the SSD in response to an operation command received from the host, and control the allocation and / or deallocation of hardware resources for the identified workload according to a budget target of the identified power state.

[0007] Additional aspects of the invention will become apparent from the following description. Attached Figure Description

[0008] Figure 1This is a diagram illustrating a data processing system according to an embodiment of the present invention;

[0009] Figure 2 This is a graph of throughput based on input workload according to another embodiment of the present invention;

[0010] Figure 3 This is a graph of throughput based on input workload according to yet another embodiment of the present invention;

[0011] Figure 4 This is a graph showing the throughput of different storage dies in an SSD according to an embodiment of the present invention, which have different power levels according to throughput.

[0012] Figure 5 This is a graph of throughput based on input workload according to yet another embodiment of the present invention;

[0013] Figure 6 This is a graph of throughput based on input workload according to yet another embodiment of the present invention;

[0014] Figure 7 This is a flowchart illustrating a method for controlling a storage device according to yet another embodiment of the present invention; and

[0015] Figure 8 This is a flowchart illustrating a method for operating a data storage system according to an embodiment of the present invention. Detailed Implementation

[0016] Various embodiments are described in more detail below with reference to the accompanying drawings. However, the invention may be embodied in different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to make this disclosure thorough and complete, and to fully convey the scope of the invention to those skilled in the art. Furthermore, references herein to “embodiment,” “another embodiment,” etc., are not necessarily directed to only one embodiment, and different references to any such phrases are not necessarily directed to the same embodiment. Throughout this disclosure, the same reference numerals refer to the same parts in the drawings and embodiments of the invention.

[0017] This invention can be implemented in various ways, including as a process; an apparatus; a system; a computer program product embodied on a computer-readable storage medium; and / or a processor, such as a processor adapted to execute instructions stored on and / or provided by memory linked to the processor. In this specification, these embodiments or any other form in which the invention may be taken can be referred to as technology. Generally, the order of steps of the disclosed process can be changed within the scope of this invention. Unless otherwise stated, components such as processors or memory described as suitable for performing tasks can be implemented as general components temporarily configured to perform tasks at a given time or manufactured as specific components for performing tasks. As used herein, the term "processor," etc., refers to one or more means, circuits, and / or processing cores adapted to process data such as computer program instructions.

[0018] The following detailed description of embodiments of the invention is provided in conjunction with the accompanying drawings, which illustrate various aspects of the invention. The invention is described in conjunction with these embodiments, but is not limited to any particular embodiment. The scope of the invention is limited only by the claims. The invention includes many alternatives, modifications, and equivalents within the scope of the claims. Numerous specific details are set forth in the following description to provide a thorough understanding of the invention. These details are provided for illustrative purposes; the invention may be practiced according to the claims without some or all of these specific details. For clarity, technical materials known in the art related to the invention have not been described in detail so as not to unnecessarily obscure the invention.

[0019] Typically, performance metrics for storage systems (such as enterprise SSDs) include throughput in megabytes per second (MB / s) or input and output (IO) operations per second (IOP), as well as latency such as average or multi-nines latency profiles. A multi-nines latency profile represents the worst command completion time among these multiple nines (e.g., 99%, 99.9%), and is often referred to as a Quality of Service (QoS) metric.

[0020] Memory systems can include memory devices that are operatively connected in parallel, such as NAND flash memory devices. Due to their parallel architecture, command processing on memory devices (e.g., data read or write operations associated with read or write commands) is typically inherently bursty. This means that significantly more commands are completed in one time slot than in any other. In this case, the performance metrics of the memory system are not constant. Therefore, a scheme capable of controlling power consumption is desired.

[0021] In fact, Environmental, Social, and Governance (ESG) requirements are increasingly driving all electronic components to become “greener” technologies. Even for SSDs, SSD performance per watt is becoming increasingly important. Therefore, the total power consumption over the SSD's lifecycle becomes significant for many different operating conditions, regardless of whether it's under high-load peak performance. Typical SSD performance metrics are throughput in MB / s, or I / O operations per second (IOP). Past focus has been on achieving optimal overall performance, not necessarily optimal performance for a given power limit.

[0022] Most (if not all) existing SSDs lack the ability to predict power and performance characteristics, let alone predict power and performance characteristics for different capacities of the same product. For example, 1TB / 2TB / 4TB / 8TB SSDs with the same product model share the same power state and the same power consumption target number regardless of their individual capacity characteristics. Moreover, the often subtle and complex power throttling logic makes the resulting performance unpredictable.

[0023] In this context, one embodiment of the present invention provides a computational algorithm that calculates (predicts) performance per watt under at least the following constraints: multiple SSD capacities, multiple power states, and multiple workloads. The computational algorithm of the present invention can be used in real time in a memory controller or before a set of storage operations, wherein prediction criteria are provided to the host so that the host considers total power consumption when sending commands to the memory device.

[0024] Figure 1 This is a diagram illustrating a data processing system according to an embodiment of the present invention.

[0025] Reference Figure 1 The data processing system may include host 5, controller 100, and memory device 200. Controller 100 and memory device 200 may configure memory system 10. For example, controller 100 and memory device 200 may include various components of memory system 10 not shown herein. Although the following description focuses on NVMe SSD-based terminology, the invention is not limited thereto but is applicable to other memory devices including different types of SSDs. In NVMe (High Speed ​​Non-Volatile Memory) SSDs, IO command queues are paired as commit queues (SQ) and completion queues (CQ).

[0026] exist Figure 1In the example embodiment shown, memory system 10 can be implemented using a high-speed non-volatile memory (NVMe) SSD, and memory device 200 can be implemented using multiple parallel memory modules, such as k parallel NAND flash memory modules NAND 1 461 to NAND 469. The NAND flash memory modules (e.g., NAND dies) 461 to 469 are connected to memory controller 100 via one or more channels.

[0027] Host 5 can exchange command "requests" and "completions" with memory system 10 via a predefined I / O command queue structure. For example, host 5 includes a submission queue (SQ) 52 and a completion queue (CQ) 54, which are pairs of I / O command queues. Host 5 can include SQ and CQ pairs according to configurations such as the number of CPU cores in controller 100, the operating system (OS), and the I / O configuration of the memory device (i.e., the memory device). Submission queue 52 can store command requests associated with the operation of memory device 200, which are submitted to memory controller 100. Completion queue 54 can receive completion notifications for command requests from controller 100, i.e., data items indicating that each command request has been completed, and stores the received completion notifications. Therefore, when corresponding completion data is placed in completion queue 54, host 5 recognizes each completion in the command request. In response to completion data being transmitted to completion queue 54, host 5 places new commands into submission queue 52 to maintain a predetermined number of commands processed by memory system 10.

[0028] Controller 100 may include a host interface (I / F) 140, a memory interface (I / F) 150, a command processing component 410, and a predictive power performance module 420. These components 410, 420 may be implemented using internal components of the control component (e.g., software (SW)). Alternatively, these components 410, 420 may be implemented using hardware (HW) components. Although in Figure 1 Although not shown, the controller 100 and memory device 200 may include a variety of other components of the memory system 10.

[0029] Host interface 140 can receive command requests from host 5. For example, the command request may be associated with an operation of memory device 200 (e.g., a read, write, or erase operation). Command processing component 410 can process the command request to execute the command request on memory device 200.

[0030] In one embodiment of the invention, the control algorithm is based on a static resource allocation table to precisely control the allocation of active resources for each workload according to the target, without the need for complex logic or hardware measurement feedback loops. Due to the varying power consumption of different memory components, workload type identification allows for different resource threshold levels.

[0031] In one embodiment of the invention, the control (or prediction) algorithm dynamically allocates or deals with active resources in a precise manner using a static hardware resource table. Among the many limiting factors / resources for throughput / power in SSDs, NAND is the primary component affecting throughput / power. Power consumption is directly proportional to the utilization rate of the NAND components. Inside an SSD, NAND comprises multiple physical packages. Each physical package includes multiple dies, typically with different operating characteristics. As SSD capacity increases, the number of dies also increases. Throughput and power are proportional to the number of dies until performance saturates due to other factors.

[0032] Depending on the allocation of hardware resources, the resulting (throughput, power) pair will increase or decrease to achieve the budgeted (throughput, power) target for a given power state on a given SSD capacity drive. As a result of an algorithm that sums up each independent power state of SSD operation, such as summing the power / throughput for each read, write, or erase event, solid-state drives of different capacities will have a predictable (throughput, power) trend in the throughput required to achieve per watt.

[0033] Figure 2 It is a graphical representation of throughput / power consumption versus host input workload.

[0034] More specifically, Figure 2 The diagram shows a conceptual representation of a general-purpose NVMe SSD (throughput, power consumption) and host input workload. Figure 2 The left side of the diagram illustrates how SSD throughput increases with increasing host input workload until the SSD reaches its saturation point (denoted here as TPmax). Similarly, the right side shows that power consumption reaches its maximum at the same time as maximum throughput. Maximum throughput is typically determined by the SSD's internal resource limitations and the type of input workload.

[0035] Because lower power is required for many mission-critical operations in different integrated environments, SSD hosts typically impose certain power limits on SSDs for specific periods of time, which are provided, for example, by the NVMe power state specification that provides the protocol for this information transmission. Figure 3 It is for a specific power state PS i The host input and (TP, Pwr) are graphically described. More specifically, as... Figure 3 As shown, the SSD operates under two different maximum power constraints, namely Pmax (hereinafter referred to as the PSO state) and PS. i,max (hereinafter referred to as PS) i (Power state) operation.

[0036] When the host applies a PS to the SSD during any of its operating periods i When limited, the throughput curve will be lower than the corresponding PS0 curve, and according to PS i The throughput curve of the SSD algorithm may have a different shape than that of the PSO curve.

[0037] Figure 4 It is a graphical description of (TP, Pwr) at different capacity drivers without power state constraints. Figure 5 In PS i Graphical description of (TP, Pwr) at different capacity drivers under constraints.

[0038] The main component of an SSD is NAND flash memory, which is the primary factor affecting throughput and power consumption. An important aspect of an SSD is its capacity, which limits how much storage space can be allocated to a user. Common characteristics of SSDs (for different storage capacities) are shown as follows... Figure 4 and Figure 5 The following trends are shown. The order of capacity size is that the storage size of device 1① is smaller than that of device 2②, and the storage size of device 2② is smaller than that of device 3③. For the same power consumption, the smaller capacity driver has higher throughput efficiency (performance per watt) because the smaller capacity driver has less overhead loss.

[0039] Figure 6 This illustrates the power state PS controlled by the algorithm of the present invention. i Improved per-watt performance in devices 2② and 3③. At the same performance level, the control algorithm maintains the power consumption difference between larger memory devices as similar to the idle power difference between smaller memory devices.

[0040] The various workloads controlled by the control algorithm of this invention for performance per watt include sequential write, sequential read, random write, and random read. Each workload defines a different set of resource utilization targets. For example, the current (or power) consumed by a NAND die during active operations such as page reads or page programming is 5 to 10 times that consumed during idle time. Based on power consumption, the target maximum power for each power state and the maximum number of active resources for each power state are determined, for example, during the design phase.

[0041] Computational Algorithm

[0042] Figure 7 This is a flowchart describing the calculation algorithm of the present invention for calculating power consumption data for each operating state of a memory device.

[0043] Consider a two-dimensional NAND resource array where static initialization is performed for each SSD capacity. (Similar allocation can be applied to other contributing resources.) An active die is defined as a die in any operational state other than an idle state (such as its page / block programming (including erase) and read operation states). Figure 7 At point 701, a static hardware resource table is generated based on different threshold numbers of page / block operation states of the memory device. As shown at point 701, the code list used to generate the static hardware resource table can be:

[0044]

[0045] Where N(i,j) is the number of active NAND chips in the SSD, i is the workload type i, W0 = sequential write, W1 = sequential read; j is the power state j, such as P0 = no throttling, P1 = step 1 throttling, P2 = step 2 throttling.

[0046] For illustration, assume the maximum number of available dies for the 512Gb / die is 128@8TB, 64@4TB, 32@2TB, and 16@1TB. For an 8TB drive, a set of static values ​​can be specified as N00 = 128, N01 = 64, N02 = 32, and N03 = 16. Then, for workload type 0, the drive will show values ​​close to (or slightly less than) the maximum unthrottled throughput of the 4TB / 2TB / 1TB drive at power states 1 / 2 / 3, respectively.

[0047] exist Figure 7 At point 703, for each new command arriving at the SSD (under any given power state ps of the BAND), the workload type is identified. In other words, at 703, the `workload_type` command can obtain the power state ps of the memory device (over a given observation interval width). For example, the `workload_type` command could be an NVMe command. As shown at 703, the type can be a specific workload type, such as, but not limited to, the following:

[0048] 0 = Sequential writing

[0049] 1 = Sequential reading

[0050] 2 = Random Read

[0051] 3 = Random write

[0052] 4 = Random mixed operations between reading and writing

[0053] Once the workload type is identified, at 705, memory device resources are allocated based on the workload type and power state required by the received command. For example, based on the workload type and power state of the received command, allocations not exceeding a maximum power threshold (such as...) are identified. Figure 6 As shown), similarly, the workload type and power state of the received command that exceed the maximum power threshold are identified for delayed allocation.

[0054] At 707, command processing that can delay the processing of received commands is shown. At 707, if the workload type and power state of the received command do not exceed the maximum power threshold (such as...), then... Figure 6 As shown), the command can be processed in the order it is received. At 707, if the workload type and power state of the received command exceed the maximum power threshold, the execution of the received command is delayed until an idle state (of the memory cell involved in the command operation) occurs. This can be achieved using... Figure 7 The following command list is shown:

[0055]

[0056]

[0057] The power state transitions between different states Pi and Pj primarily occur due to power budget requirements indicated by the host application, if not always. The resulting power state transitions will be transmitted for each SSD specification, such as the NVMe specification.

[0058] The method described above can be used to operate memory systems including servers and multiple SSDs. This method can be applied to NAND flash-based SSDs (as described above) and is particularly advantageous for memory devices with multiple power states, and especially for hardware resources (e.g., in SSD NVMe) that exhibit significantly different power consumption between idle and active states. This method is particularly beneficial for hardware resources with dynamically adjustable target (power, performance) standards.

[0059] Figure 8 This is a flowchart illustrating a method for operating a data storage system comprising a host and at least one solid-state drive (SSD). At 801, the method for operating the data storage system identifies a workload associated with the SSD (e.g., a workload associated with an operation command received from the host). At 803, the method (e.g., based on the identified workload) identifies the power state of the SSD. At 805, the method controls the allocation and / or unallocation of hardware resources for the identified workload according to a budget target for the power state.

[0060] In this method for operating a data storage system, a set of static hardware resource tables can be generated for each power state of the SSD based on multiple operation commands. The power state can be identified by recognizing the power state associated with the operation commands received from the host from this set of static hardware resource tables.

[0061] In this method for operating a data storage system, when the number of dies required by an operation command exceeds the maximum active die limit of the SSD, the operation command is queued for execution until the active die becomes idle. Conversely, when the number of dies required by an operation command is less than the maximum active die limit of the SSD, resources are allocated for the operation command and the operation command is executed.

[0062] In this method for operating a data storage system, the maximum power consumption (e.g., a maximum power consumption value) of the operation command received from the host can be provided to the host, and the maximum power consumption of the operation command received from the host can be predicted based on the workload of the operation command and the power state of the SSD. In this method, the maximum hardware allocation for the SSD can be predicted based on the identified workload of the operation command and the identified power state of the SSD.

[0063] In this method of operating a data storage system, hardware resources can be allocated among different memory devices in an SSD having multiple power states, wherein the multiple power states have different power consumption between an idle power state and an active power state including at least one of a read state, a programmable state, and an erase state.

[0064] Accordingly, in one embodiment of the present invention, a system for operating a data processing system is provided, the system comprising: a host; a memory storage device; and a controller included in the memory storage device, the controller being configured to: identify a workload associated with an SSD, identify the power state of the SSD in response to an operation command received from the host, and control the allocation and / or deallocation of hardware resources for the identified workload based on a budget target of the identified power state.

[0065] The controller in this system can be configured to generate a set of static hardware resource tables for each power state of the SSD based on multiple operational commands. The controller can be configured to identify the power state associated with the operational commands received from the host from the set of static hardware resource tables.

[0066] The controller in this system can be configured to queue operation commands for execution when the number of dies required by the operation command exceeds the maximum active die limit of the SSD, until the active die becomes idle. The controller can also be configured to allocate resources for and execute operation commands when the number of dies required by the operation command is less than the maximum active die limit of the SSD.

[0067] In this system, the maximum power consumption of the SSD is provided to the host based on the operation commands received from the host. The controller can be configured to predict the maximum power consumption of the operation commands received from the host based on the workload of the operation commands and the power state of the SSD. The controller can also be configured to predict the maximum hardware allocation of the SSD based on the workload of the operation commands and the power state of the SSD. Finally, the controller can be configured to allocate hardware resources among different memory devices in SSDs with multiple power states.

[0068] In this system, there can be multiple power states with different power consumptions between the idle or non-operating power state and the default power state PS0 with no power throttling or minimum power throttling.

[0069] Although the foregoing embodiments have been shown and described in considerable detail for clarity and ease of understanding, the invention is not limited to the details provided. Those skilled in the art will appreciate from the foregoing disclosure that many alternative ways of carrying out the invention exist. Therefore, the disclosed embodiments are illustrative and not restrictive. The invention is intended to include all modifications and substitutions recognized by those skilled in the art.

Claims

1. A method of operating a system comprising a host and at least one solid-state drive, i.e., an SSD, the method comprising: Identify the workload associated with the SSD; Identify the power state of the SSD in response to operation commands received from the host; as well as Based on the budget target of the identified power state, control the allocation and / or deallocation of hardware resources for the identified workload. If the power status of the workload corresponding to the received command does not exceed the budget target, the operation commands are processed in the order of the received operation commands. If the power status of the workload corresponding to the received command exceeds the budget target, the execution of the received command is delayed.

2. The method of claim 1, wherein, Identifying workloads involves generating a set of static hardware resource tables for each power state of the SSD based on multiple operation commands.

3. The method according to claim 2, wherein, Identifying power states includes: Identify the power state associated with the operation command received from the host from the set of static hardware resource tables.

4. The method according to claim 3, wherein, Control allocation includes: When the number of dies required for an operation command exceeds the maximum active die limit of the SSD, the operation command is queued for resource allocation until any active die becomes idle.

5. The method according to claim 4, wherein, Control allocation includes: When the die required for the operation command is less than the maximum active die limit of the SSD, resources are allocated for the operation command and the operation command is executed.

6. The method of claim 1, further comprising: Based on the operation command received from the host, the maximum power consumption of the SSD is provided to the host.

7. The method of claim 6, further comprising: Based on the workload of the operation command and the power state of the SSD, predict the maximum power consumption of the operation command received from the host.

8. The method of claim 6, further comprising: Based on the workload of the operation command and the power state of the SSD, the maximum hardware allocation of the SSD is predicted.

9. The method of claim 3, further comprising: Hardware resources are allocated among the different memory devices in the SSD, which have multiple power states.

10. The method according to claim 9, wherein, The various power states have different power consumption between the idle or non-operating power state and the default power state with minimum power throttling.

11. A system for operating a data processing system, comprising: Host; Memory storage device; as well as The controller, included in the memory storage device, wherein the controller: Identify the workloads associated with solid-state drives, i.e., SSDs; Identify the power state of the SSD in response to operation commands received from the host; as well as Based on the budget target of the identified power state, control the allocation and / or deallocation of hardware resources for the identified workload. If the power status of the workload corresponding to the received command does not exceed the budget target, the operation commands are processed in the order of the received operation commands. If the power status of the workload corresponding to the received command exceeds the budget target, the execution of the received command is delayed.

12. The system according to claim 11, wherein, The controller generates a set of static hardware resource tables for each power state of the SSD based on multiple operation commands.

13. The system according to claim 12, wherein, The controller identifies the power state associated with the operation command received from the host from the set of static hardware resource tables.

14. The system according to claim 13, wherein, The controller: When the number of dies required for an operation command exceeds the maximum active die limit of the SSD, the operation command is queued for resource allocation until any active die becomes idle.

15. The system according to claim 14, wherein, The controller: When the die required for the operation command is less than the maximum active die limit for the SSD, resources are allocated for the operation command and the operation command is executed.

16. The system according to claim 11, wherein, The maximum power consumption of the SSD is provided to the host based on the operation command received from the host.

17. The system according to claim 16, wherein, The controller predicts the maximum power consumption of the operation command received from the host based on the workload of the operation command and the power state of the SSD.

18. The system according to claim 16, wherein, The controller predicts the maximum hardware allocation for the SSD based on the workload of the operation command and the power status of the SSD.

19. The system according to claim 13, wherein, The controller allocates hardware resources among different memory devices in the SSD, which have multiple power states.

20. The system according to claim 19, wherein, The various power states have different power consumption between the idle or non-operating power state and the default power state with minimum throttling.

Citation Information

Patent Citations

  • Data storage device and operating method thereof

    CN110286847A

  • Storage device and method for adaptive sleep mode transitions

    US20190129636A1