Storage device, storage system and throttling method thereof

Patent metadata

Field	Value
Publication number	US-2024168674-A1
Application number	US-202318127922-A
Country	US
Kind code	A1
Filing date	Mar 29, 2023
Priority date	Nov 18, 2022
Publication date	May 23, 2024
Grant date	—

Abstract

A throttling method for a storage device is provided. The throttling method includes: receiving a write command from a host; identifying, using a first machine learning model, a throttling delay time; transmitting a completion message to the host according to the throttling delay time; collecting weights of the first machine learning model and performance information of the storage device corresponding to the weights; learning the weights and the performance information to generate an objective function indicating a relationship between the weights and the performance information using a second machine learning model of a weight learning device; selecting a weight corresponding to a maximum performance using the objective function; and updating the first machine learning model with the weight.

First claim

What is claimed is:

1 . A throttling method for a storage device, the throttling method comprising:
receiving a write command from a host; identifying, using a first machine learning model, a throttling delay time; transmitting a completion message to the host according to the throttling delay time; collecting weights of the first machine learning model and performance information of the storage device corresponding to the weights; learning the weights and the performance information to generate an objective function indicating a relationship between the weights and the performance information using a second machine learning model of a weight learning device; selecting a weight corresponding to a maximum performance using the objective function; and updating the first machine learning model with the weight.

2 . The throttling method of claim 1 , wherein the identifying the throttling delay time comprises receiving state information and workload information of the storage device.

3 . The throttling method of claim 2 , wherein the state information or the workload information comprises any one or any combination of a number of write buffers in use, a write ratio, a queue depth, and a write amplification factor of the storage device.

4 . The throttling method of claim 1 , further comprising identifying the performance information using a weighted average of throughput, quality of service, and consistency.

5 . The throttling method of claim 1 , wherein further comprising training the second machine learning model by applying a Bayesian Optimization Surrogate Model for generating the objective function.

6 . The throttling method of claim 5 , wherein the objective function maps the weights as inputs and the performance information as outputs.

7 . The throttling method of claim 1 , wherein an initial weight of the first machine learning model is provided as a random value.

8 . A storage device comprising:
a non-volatile memory device; a buffer memory configured to buffer write data corresponding to a write command before writing the write data to the non-volatile memory device; one or more processors configured to:
identify a throttling delay time using a first machine learning model;
collect weights of the first machine learning model and performance information of the storage device; and
generate an objective function indicating a relationship between the weights and the performance information using a second machine learning model.

9 . The storage device of claim 8 , wherein the one or more processors are further configured to control the first machine learning model to identify the throttling delay time using state information and workload information of the storage device.

10 . The storage device of claim 9 , wherein the state information of the storage device or the workload information indicates any one or any combination of a number of write buffers in use, a write ratio, a queue depth, and a write amplification factor of the storage device.

11 . The storage device of claim 8 , wherein the one or more processors are further configured to:
collect the weights of the first machine learning model and the performance information of the storage device; generate the objective function indicating the relationship between the weights and the performance information; and identify selected weights corresponding to target performance using the objective function.

12 . The storage device of claim 11 , wherein the one or more processors are further configured to map the weights as inputs of the objective function and the performance information as outputs of the objective function.

13 . The storage device of claim 12 , wherein the one or more processors are further configured to train the second machine learning model by applying a Bayesian Optimization Surrogate Model for generating the objective function.

14 . The storage device of claim 11 , wherein the target performance corresponds to maximum performance of the storage device.

15 . The storage device of claim 8 , wherein the one or more processors are further configured to identify the performance information of the storage device using a weighted average of throughput, quality of service, and consistency.

16 . A storage system, comprising:
a storage device comprising one or more first processors configured to control the storage device to:
identify a throttling delay time corresponding a write command from a host; and
transmit a completion message to the host according to the throttling delay time; and
a weight learning device comprising one or more second processors configured to control the weight learning device to:
collect weights of a first machine learning model and performance information of the storage device corresponding to the weights; and
generate an objective function indicating a relationship between the weights and the performance information using a second machine learning model.

17 . The storage system of claim 16 , wherein the first machine learning model is configured to identify the throttling delay time based on state information and workload information of the storage device.

18 . The storage system of claim 16 , wherein the second machine learning model is trained by applying a Bayesian Optimization Surrogate Model for generating the objective function.

19 . The storage system of claim 16 , wherein the one or more second processors are further configured to repeat training of the second machine learning model until the performance information is equal to or higher than a reference level.

20 . The storage system of claim 16 , wherein the performance information of the storage device is identified using a weighted average of throughput, quality of service, and consistency.

Assignees

• Samsung Electronics Co Ltd

Inventors

• Jung Kibeen
• Lee Han Kyoo
• Kim Byeonghui
• Oh Hyunkyo
• Jang Sungmin

Classifications

• G06N20/00

  Machine learning · CPC title

• G06F3/0679

  Non-volatile semiconductor memory device, e.g. flash memory, one time programmable memory [OTP] · CPC title

• G06F3/0659 Primary

  Command handling arrangements, e.g. command buffers, queues, command scheduling · CPC title

• G06F3/0656

  Data buffering arrangements · CPC title

• G06F3/0653 Primary

  Monitoring storage devices or systems · CPC title