KR20200055898A - Low-power raid scheduling method for distributed storage application - Google Patents

Abstract

Disclosed is a low-power redundant array of inexpensive disks (RAID) scheduling method for distributed a storage application. According to an embodiment, a RAID scheduling method includes the steps of: receiving a chunk of data from an initiator server, by the distributed storage application, generating a parity block using the CPU on the target storage server, and separately stripping the data block using an SSD and the parity block using an HDD; and converting the power mode of the SDD and the HDD, by an energy-aware encoding scheduler, from an active mode to an idle mode or a standby mode.

Description

LOW-POWER RAID SCHEDULING METHOD FOR DISTRIBUTED STORAGE APPLICATION}

The following embodiments relate to a low-power RAID scheduling technique for a distributed storage application, and more particularly, to a power mode scheduling technique for reducing energy consumption in a distributed storage application.

The recent trend in storage application development suggests that researchers need to make a breakthrough in distributed storage systems in terms of energy consumption. In distributed storage applications (distributed storage applications), RAID (Redundant Array of Inexpensive Disks) requires high energy consumption because data and parity blocks must be striped to the storage node. RAID empowers storage applications to tolerate node failures. First, RAID divides a chunk of data into blocks of data at the file system level. Then, RAID creates a redundant data block from the data block. Finally, RAID performs concurrent physical writes at the kernel level.

Table 1 shows a comparison between scheduling methods developed for distributed storage applications.

[Table 1]

Table 1 shows the I / O levels and DPM processes of erasure codes, storage networks, disk arrays and dynamic power management (DPM). From the perspective of, the contents of the storage application proposed in this embodiment are compared with the contents of the existing storage application.

In the existing studies of Son et al. (Non-patent document 1), Zhu et al. (Non-patent document 2), and Pirahandeh et al. (Non-patent document 3), the DPM scheduler is used for a single logical I / O operation or multiple logical I / O operations. Applies. Previous studies suggested two types of DPM: off-line and on-line. Offline DPM knows the length of the storage node's idle period in advance, whereas online DPM does not. In addition, from the perspective of the DPM process type, the existing DPM schedulers proposed by Son et al. (Non-patent document 1) and Pirahandeh et al. (Non-patent document 3) use an online process, and proposed by Zhu et al. (Non-patent document 2). The DPM scheduler uses online or offline courses.

 Woo, S.S., Kandemir, M .: Runtime system support for softwareguided disk power management. In: Processing of the Cluster Computing, pp. 139-148 (2007) . Zhu, Q., David, F.M., Devaraj, C.F., Li, Z., Zhou, Y., Cao, P .: Reducing energy consumption of disk storage using power-aware cache management. In: Processing of the IEEE on Software (2004)  Pirahandeh, M., Kim, H .: Reliable energy-aware SSD based RAID-6 system. In: Proceedings of the FAST Conference in Storage Systems, San Jose, USA (2012)

Embodiments describe a low-power RAID scheduling technique for distributed storage applications, and more specifically, provide a power mode scheduling technique for reducing energy consumption in distributed storage applications.

The embodiments pipeline physical data blocks using an online DPM process to utilize a high transmission rate in the case of an encoder, and parity blocks are subsequently pipelined using an offline DPM process, thereby enabling online and offline DPM. It is to provide a low-power RAID scheduling technique for distributed storage applications that reduces energy consumption in distributed storage applications by using all methods.

Embodiments are low-power RAID scheduling scheme of a distributed storage application that can reduce energy consumption by allowing the storage application to switch the power mode of the storage node from an active state to an idle state or a standby state without performing any I / O operations. To provide.

According to an exemplary embodiment, a RAID (Redundant Array of Inexpensive Disks) scheduling method, a distributed storage application receives a data chunk from an initiator server, generates a parity block using a CPU in a target storage server, and uses SSD data Separately striping the parity blocks using blocks and HDDs; And the energy-aware encoding scheduler comprising switching the power modes of the SDD and the HDD from an active mode to an idle mode or a standby mode.

The energy-aware encoding scheduler converts the power modes of the SDD and the HDD from an active mode to an idle mode or a standby mode, and pipes the physical data blocks using an online dynamic power management (DPM) method. step; And pipelining the parity blocks using an offline DPM method, and reducing energy consumption of the distributed storage application using the online DPM method and the offline DPM method.

According to another embodiment of the RAID (Redundant Array of Inexpensive Disks) scheduling system, in order to reduce energy consumption of a distributed storage application that separately stripes a data block using an SSD and a parity block using an HDD, And an energy-aware encoding scheduler that switches the power mode of the SDD and the HDD from an active mode to an idle mode or a standby mode.

The energy-aware encoding scheduler includes: an online dynamic power management (DPM) unit that physically pipelines the data blocks; And an offline DPM unit for pipelining the parity blocks, and using the online DPM unit and the offline DPM unit to reduce energy consumption of a distributed storage application.

In addition, a data chunk is received from an initiator server, and the target storage server generates a parity block using a CPU, and separately stripes the data block using the SSD and the parity block using the HDD. It may further include a distributed storage application.

According to embodiments, in the case of an encoder, the physical data blocks are pipelined using an online DPM process to use a high transmission rate, and the parity blocks are subsequently pipelined using an offline DPM process, thereby allowing online and It is possible to provide a low-power RAID scheduling technique for distributed storage applications that reduces energy consumption of distributed storage applications by using all offline DPM methods.

According to embodiments, the storage application does not perform any I / O operation, and the power mode of the storage node can be switched from an active state to an idle state or a standby state, thereby reducing energy consumption. Scheduling techniques can be provided.

1 is a diagram illustrating a DPM-based scheduling method in a conventional distributed storage application.
2 is a flowchart illustrating a RAID scheduling method according to an embodiment.
3 is a block diagram illustrating a RAID scheduling system according to an embodiment.
4 is a diagram for explaining a RAID scheduling method based on DPM in a distributed storage application according to an embodiment.
5 is a diagram illustrating an example of an encoding power management model according to an embodiment.
6 is a diagram for explaining an example of a decoding power management model according to an embodiment.

Hereinafter, embodiments will be described with reference to the accompanying drawings. However, the described embodiments may be modified in various other forms, and the scope of the present invention is not limited by the embodiments described below. In addition, various embodiments are provided to more fully describe the present invention to those skilled in the art. The shape and size of elements in the drawings may be exaggerated for a more clear description.

In the following embodiment, power mode scheduling for reducing energy consumption in distributed storage applications is proposed. The proposed distributed storage application may be composed of an initiator sever, a target storage server, and an iSCSI high-speed storage network. The proposed distributed storage application uses a CPU to generate parity, and can strip parity and data blocks on the target storage server. The energy aware encoding scheduler allows the storage application to switch the power mode of the storage node from active to idle or redundant without performing any I / O operations.

라고 가정한다. i 번째 데이터 청크는 nw 개의 데이터 블록으로 구성되며,

이다. i 번째 패리티 청크는 패리티 블록들로 구성되며,

이다. 여기서, mw는 스트라이프의 개수를 나타낸다. Distributed RAID storage applications designed to ensure reliability create coded data nodes called "parity nodes" using various erasure coding methods to provide node fault tolerance. do. There are various types of erasure codes such as Reed-Solomon (RS) codes and Single Parity Check (SPC) codes, which are defined by coding schemes. In a distributed RAID storage application, workload D consists of k data chunks, i.e.

Is assumed. The i th data chunk consists of nw data blocks,

to be. The i-th parity chunk consists of parity blocks,

to be. Here, mw represents the number of stripes.

Table 2 shows a pseudo code of a parity generator using a Cauchy RS code.

[Table 2]

라고 가정한다. 여기서, 생성자 행렬(generator matrix)

은 wn x wn 코드 워드를 갖는 항등 행렬 I 및 m x n 코딩 블록을 갖는 코딩 행렬 X =

로 구성된다. XOR 연산은 스트라이프의 데이터 코드 워드들 간에 수행된다. CPU 코어를 사용하는 분산형 스토리지 어플리케이션(SA1, SA2, SA4 및 SA5)에서, i 번째 스트라이프에서 패리티 코드 워드 P_i를 생성하기 위한 XOR 연산의 시퀀스는 다음 식과 같이 나타낼 수 있다.In this laser code,

Is assumed. Here, the generator matrix

Is the identity matrix I with wn x wn code words and the coding matrix X = with m x n coding blocks

It consists of. The XOR operation is performed between stripes of data code words. In a distributed storage application (SA1, SA2, SA4, and SA5) using a CPU core, a sequence of XOR operations for generating the parity code word P _i in the i-th stripe can be expressed as the following equation.

[Equation 1]

First, an existing energy-aware I / O scheduling method for a distributed storage application will be described.

1 is a diagram illustrating a DPM-based scheduling method in a conventional distributed storage application.

Referring to FIG. 1, a logical write scheduling method (k = 2, n = 2, m = 1, w = 2) based on DPM in a conventional distributed storage application is shown, and two slave data nodes (two In a conventional RAID 5 storage application using one SSD) and one slave parity node (one HDD), a sequence diagram of an existing energy-aware I / O scheduling method for a distributed storage application is shown. The logical write operation delay (LW) of the existing RAID level 5 can be expressed as the following equation.

[Equation 2]

Table 3 shows the notation list.

[Table 3]

As shown in Table 3, Spin up, Ch, ABC, ABR, XOR, SR, PW, and Spin dawn each change the disk power mode from standby / idle mode to active mode, Splitting data into chunks, allocating blocks to main memory, allocating blocks from or to registers, performing XOR operations using the parity computing scheduler, and using the system kernel Denotes delays by reading blocks from main memory, physically writing blocks to disk using a RAID device driver, and switching disk power mode from active mode to standby / idle mode.

Since the next command can be sent after a Write Completion (WC) response to the target storage server, the LW delay is the elapsed time between the command of the iSCSI initiator and the response to completion of the command (t _wc) ). Finally, logical writes can be handled online or offline regardless of the type of storage device in the node array.

2 is a flowchart illustrating a RAID scheduling method according to an embodiment.

Referring to FIG. 2, in a RAID scheduling method according to an embodiment, a distributed storage application receives a data chunk from an initiator server, generates a parity block using a CPU in a target storage server, and a data block using an SSD And striping the parity blocks using the HDD separately (S110), and the energy-aware encoding scheduler converting the power mode of the SDD and HDD from an active mode to an idle mode or a standby mode (S120). Can be achieved.

The energy-aware encoding scheduler converts the power modes of the SDD and HDD from an active mode to an idle mode or a standby mode (S120), by using an online dynamic power management (DPM) method to pipeline physical data blocks. Step S121, and pipelining the parity blocks using an offline DPM method (S122), may reduce energy consumption of the distributed storage application using an online DPM method and an offline DPM method.

A RAID scheduling method according to an embodiment will be described as an example below.

Dynamic Power Management (DPM) methods make power-mode-related decisions based on information available at run time (online, online) or before (offline, offline). Get off. The present embodiments provide an energy-aware RAID scheduling method that reduces energy consumption of distributed storage applications using both online and offline DPM methods. The proposed Energy-Aware Redundant Array of Inexpensive Disk (RAID) scheduling method allows the storage application to stripe data and parity blocks individually without performing any physical I / O operations, and power mode is idle in active mode. ) It is different from the existing RAID methods in that it switches to the standby mode or standby mode.

The RAID scheduling method according to an embodiment may be described in more detail through a RAID scheduling system according to an embodiment.

3 is a block diagram illustrating a RAID scheduling system according to an embodiment.

Referring to FIG. 3, the RAID scheduling system 100 according to an embodiment may include an energy-aware encoding scheduler 120, and the energy-aware encoding scheduler 120 includes an online DPM unit 121 and an offline DPM unit It can be made, including 122. According to an embodiment, the RAID scheduling system 100 may further include a distributed storage application 110.

In step S110, the distributed storage application 110 receives the data chunk from the initiator server, generates a parity block using a CPU in the target storage server, and a data block using an SSD and a parity block using an HDD Can be striped individually.

In step S120, the energy-aware encoding scheduler 120 may switch the power modes of the SDD and HDD from an active mode to an idle mode or a standby mode.

Here, the energy-aware encoding scheduler 120 may include an online DPM unit 121 and an offline DPM unit 122.

In step S121, the online DPM unit 121 may pipelin the physical data blocks using the online DPM method.

In step S122, the offline DPM unit 122 may pipeline the parity blocks using the offline DPM method.

In the case of the storage application proposed in this embodiment, the DPM scheduler enables efficient use of the storage device by changing the power mode to a low power mode without performing any I / O operations by the storage system.

This hybrid storage application is achieved through a combination of SSD and HHD when arranging nodes. In the proposed distributed storage application 110, the encoder pipelines physical data blocks using an online DPM process to utilize a high transmission rate, while parity blocks pipeline later using an offline DPM process. do. This will allow the proposed distributed storage application 110 to perform individual DPM methods on data nodes using SSDs and parity nodes using HDDs.

Existing DPM schedulers are applied to a single logical I / O operation or multiple logical I / O operations (workload, workload), while the proposed DPM scheduler is applied to a physical I / O operation. As a result of the experiment, it was confirmed that the average energy consumption is reduced by 26-36% when the proposed storage application is compared with the existing storage application.

Hereinafter, a RAID scheduling method and a system according to an embodiment will be described in more detail.

First, a distributed storage application according to one embodiment will be described.

The proposed energy-aware distributed storage application (simply referred to as a distributed storage application or storage application) (SA3) can be configured including an initiator server, a target storage server, and an iSCSI high-speed storage network, and distributed storage It can reduce the energy consumption of the system.

In the initiator server, a data chunking module classifies the file into a number of data chunks, which are then transferred over the storage network to the target storage server using the iSCI protocol.

In the target storage server, one chunk of data is divided into a number of data blocks stored in the CPU main memory. The parity generator generates parity blocks using a parity compute function and a memory allocator function.

By placing a parity generator in the target storage server, the I / O performance of the proposed distributed storage application can be compared with the I / O performance of the existing storage application. The reason is that in all the existing storage applications (SA1, SA2, SA4, SA5), the parity generator was present in the same storage server.

The proposed energy-aware encoding scheduler uses a power-mode-switching function and a disk (SSD and HHD) energy profiler to switch the power mode of the data disk. The energy aware encoding scheduler is used to reduce power consumption by switching the SSD power mode between active and idle modes, or by switching the HDD power mode between active and standby modes.

Finally, the I / O redirector can write data and parity blocks to hybrid storage using an energy-aware scheduler method.

4 is a diagram for explaining a RAID scheduling method based on DPM in a distributed storage application according to an embodiment.

Referring to FIG. 4, a DPM-based logical write RAID scheduling method (k = 1, n = 2, m = 1, w = 2) of a storage application (SA3) relates to two data nodes (two SSDs) and One parity node (one HHD) can be used.

The logical write operation delay LW of the proposed distributed storage application may be defined as the following equation.

[Equation 3]

[Equation 4]

지연은 iSCSI 이니시에이터(initiator)의 명령과 데이터 완성 반응 간의 경과 시간(

)으로 정의될 수 있다. 다음 명령이 목표 스토리지 서버의 데이터 쓰기 완성(Data Write Completion, DWC) 반응 이후에 전송될 수 있기 때문에,

지연은 데이터 완성 반응 명령과 패리티 완성 반응 간의 경과 시간(

)으로 정의될 수 있다.here,

The delay is the elapsed time between the command of the iSCSI initiator and the data completion reaction (

). Because the following command can be sent after the Data Write Completion (DWC) response of the target storage server,

The delay is the elapsed time between the data completion response command and parity completion response (

).

Hereinafter, an energy-aware distributed RAID decoder will be described.

Table 4 shows a distributed RAID decoder capable of recovering m disk failures.

[Table 4]

가 m을 초과하면, 복구될 수 없는 완전한 실패(disaster failure)이다. 라인 (5-13)에서,

번의 디스크 실패가 발생하면, 알고리즘은 최대 mw 개의 실패된 데이터 코드 워드들을 복구한다. 라인 (6)에서, 분산형 RAID 디코더는 패리티 노드에 위치하는 패리티 디스크들의 회전을 빠르게 한다. 라인 (7)에서, 분산형 RAID 디코더는 대응하는 생존 데이터 및 패리티 코드 워드를 ds에 할당한다. 라인 (8)에서, 분산형 RAID 디코더는 패리티 노드에 위치한 패리티 디스크들의 회전을 느리게 한다. 라인 (9-10)에서, RAID 디코더는 역 코딩 코드 워드(inverted coding code words)

를 저장하며, 실패된 데이터 코드 워드(failed data code words)에 대응하는 열들이 제거된다. 라인 (11)에서, 패리티 계산 알고리즘(parity computation algorithm)을 사용하여, 분산형 RAID 디코더는 역 코딩 코드 워드

와 생존 데이터 및 패리티 코드 워드 ds 간의 행렬 곱을 수행함으로써 실패된 데이터 코드 워드를 복구한다. 라인 (12)에서, 생존 데이터 코드 워드 및 복구된 데이터 코드 워드

은

을 복구하기 위해 합쳐진다. 마지막으로, 라인 (13)에서 복구된 데이터 청크

는 메인 메모리에서 호스트로 읽혀진다.Referring to Table 4, in line (1-3), the distributed RAID decoder speeds up the rotation of data disks located in the data node (Spins up), reads a chunk of data from n data disks, and finally rotates the data disks. Slow down. On line 4, the number of data disk failures

If m exceeds m , it is a complete failure that cannot be repaired. On line (5-13),

If one disk failure occurs, the algorithm recovers up to mw failed data code words. At line 6, the distributed RAID decoder speeds up the rotation of parity disks located at the parity node. At line 7, the distributed RAID decoder assigns the corresponding survival data and parity code words to ds . At line 8, the distributed RAID decoder slows the rotation of the parity disks located at the parity node. In line 9-10, the RAID decoder is inverted coding code words

And the columns corresponding to the failed data code words are removed. At line 11, using a parity computation algorithm, the distributed RAID decoder is inverse coded code word.

The failed data code word is recovered by performing matrix multiplication between and survival data and parity code word ds . In line 12, the survival data code word and the recovered data code word

silver

To recover them. Finally, the chunk of data recovered in line (13)

Is read from main memory to the host.

In a RAID storage system, data must be stored with parity data stored across many storage devices so that even if the storage device fails, there is still enough data to recover the disk. It is important to develop an energy-saving storage system based on the RAID structure.

Disk Power Management (DPM) should save energy by putting disks into a low-power mode whenever possible without affecting performance. As soon as the I / O operation is complete, DPM determines whether the disk should remain in idle mode. However, it is more advantageous to rotate the disk slowly in a low power standby mode, since the idle period must consider the cost of speeding up or slowing down the disk rotation. Therefore, the disk rotates slowly immediately after the current request is serviced, and quickly rotates to the active mode at the time for the next request. Besides, it is better to be in idle mode after the current request is completed.

In addition, the DPM method must make a decision using only partial information. Online DPM must make a decision on resource costs before all input to the system is available. Therefore, the online DPM does not know the length of the idle period until the end. However, the offline DPM knows the length of the idle period in advance.

Here, it is possible to propose a new DPM that encodes and decodes physical data blocks for disk power management. The proposed DPM allows online DPM to process data SSD and offline DPM to process parity HDD. The total energy cost E (t) of the CPU-based encoding can be calculated based on the following equation.

[Equation 5]

[Equation 6]

[Equation 7]

,

및

는 각각 유휴(idle)/대기(standby) 시간, 패리티 생성 시간 및 읽기/쓰기 시간을 나타낸다. coding, active 및 dile/standby 모드를 위한 전력 소비는 각각

,

및

라고 표현되며, 각 CPU 사이클에 대한 전력 소비는

라고 표현된다. 데이터 디스크, 코딩 디스크 및 CPU 코어의 개수는 각각 n, m,

라고 표현된다.here,

,

And

Denotes idle / standby time, parity generation time, and read / write time, respectively. Power consumption for coding, active and dile / standby modes are respectively

,

And

And the power consumption for each CPU cycle

Is expressed. The number of data disks, coding disks and CPU cores is n, m, respectively.

Is expressed.

는 활성 모드에서 유휴/대기 모드로 디스크 회전을 느리게 하는 데 요구되는 시간이고,

는 유휴/대기 모드에서 활성 모드로 디스크 회전을 빠르게 하는 데 요구되는 에너지이다. [수학식 7]에 나타난 바와 같이, 스토리지 시스템의 스루풋 Thr은 목표 스토리지 서버 측의 인코딩 시간에 대한 데이터 크기

의 비율이다. 따라서, 인코딩 시간은 유휴(idle) 시간, 대기(standby) 시간, 코딩(coding) 시간 및 활성(active) 시간의 합으로 계산된다.In addition,

Is the time required to slow the disk rotation from active mode to idle / standby mode,

Is the energy required to accelerate the disk rotation from idle / standby mode to active mode. As shown in [Equation 7], the throughput Thr of the storage system is the data size for the encoding time of the target storage server.

It is the ratio of. Accordingly, the encoding time is calculated as the sum of idle time, standby time, coding time, and active time.

Below we analyze the energy perception model.

5 is a diagram illustrating an example of an encoding power management model according to an embodiment.

Referring to FIG. 5, an example of three RAID logical write (encoding) power management models is shown, (a) is a power management model according to an embodiment, and (b) is a conventional naive model, (c) shows the model of the existing Pirahandeh (k = 2, w = 2, n = 2, m = 1).

6 is a diagram for explaining an example of a decoding power management model according to an embodiment.

Referring to FIG. 6, an example of three RAID logical degraded read (decoding) power management models is shown, (a) is a power management model according to an embodiment, and (b) is a conventional naive (naive) model, and (c) represents the existing model of Pirahandeh (k = 2, w = 2, n = 2, m = 1).

), CPU를 사용해 패리티가 생성되거나(

), 코드 워드들이 스토리지 장치에 쓰이거나(

) 할 때, 전력 모드가 전환될 수 있도록 한다.As shown in Figures 5 and 6, three RAID power management models are used to measure energy flow. SSD has power modes of ON, OFF, active and idle, and HDD has power of sleep (drive is off), standby (drive is in low power mode) and active (normal operation) It has a mode. RAID power management devices require encoders or decoders to read data code words from storage devices (

), Or parity is generated using the CPU (

), Or code words are written to the storage device (

), The power mode can be switched.

5A and 6A show the energy flow of the encoding and decoding process using the RAID power management model SA3 according to an embodiment, respectively. To reduce energy consumption, the proposed power management model shifts disk power from idle / standby mode to active mode as each read / write operation occurs (rotation speed increases). The proposed power management model converts the disk power from active mode to idle / standby mode after each read / write operation is performed (rotation speed decreases).

5B and 6B show a naive RAID power management model proposed by Son et al. (Non-patent document 1) (SA1) and Zhu et al. (Non-patent document 2) (SA2), respectively. The naive model is designed based on a CPU-based parity generation algorithm. To reduce energy consumption, the naive model converts disk power from idle / standby mode to active mode before the encoding or decoding process begins (increasing rotational speed). The naive model converts the disk power from active mode to idle / standby mode after the encoding or decoding process is completed (lower rotation speed).

5c and 6c show the RAID power management model proposed by Pirahandeh et al. (Non-Patent Document 3). This model is designed based on the CPU-based parity generation algorithm. To reduce energy consumption, models such as Pirahandeh convert SSD power from idle to active mode (increased rotational speed) before the encoding process begins. This model converts SSD power from active to idle mode after the encoding process is complete (lower speed). HHD power will only be in active mode when the encoder performs a parity write operation. During the decoding process, if SSD2 fails, the surviving SSD power will be in active mode, and if the failed data is recovered, the failed SSD will be in active mode. However, this model puts the HDD power into active mode when the decoder recovers the failed data by accessing the surviving SSD and the parity HDD.

The energy-aware scheduling system consists of a disk energy profiler and power-mode-switching functions. The disk energy profiler sets the power mode profile of the storage device. The power mode switching function will also convert storage device power to idle / standby power mode based on the proposed RAID power management model when the I / O operation is completed.

As a result, the average encoding energy consumption of the distributed storage application SA3 according to an embodiment is lower than the energy consumption of each of the existing storage systems of SA1, SA2, SA4, and SA5. In addition, the distributed storage application SA3 according to an embodiment has higher encoding and decoding performance than existing storage systems of SA1, SA2, SA4, and SA5.

The device described above may be implemented with hardware components, software components, and / or combinations of hardware components and software components. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor (micro signal processor), a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may run an operating system (OS) and one or more software applications running on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a processing device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and / or a plurality of types of processing elements. It can be seen that may include. For example, the processing device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible.

The software may include a computer program, code, instruction, or a combination of one or more of these, and configure the processing device to operate as desired, or process independently or collectively You can command the device. Software and / or data may be interpreted by a processing device, or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. Can be embodied in The software may be distributed over networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, or the like alone or in combination. The program instructions recorded in the medium may be specially designed and configured for the embodiments or may be known and usable by those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs, DVDs, and magnetic media such as floptical disks. -Hardware devices specifically configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, etc., as well as machine language codes produced by a compiler.

As described above, although the embodiments have been described by a limited embodiment and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques are performed in a different order than the described method, and / or the components of the described system, structure, device, circuit, etc. are combined or combined in a different form from the described method, or other components Alternatively, even if substituted or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (5)

In the RAID (Redundant Array of Inexpensive Disks) scheduling method,
The distributed storage application receives a chunk of data from the initiator server, generates a parity block using a CPU in the target storage server, and stripes the data block using an SSD and the parity block using an HDD individually. step; And
The energy-aware encoding scheduler converts the power modes of the SDD and the HDD from an active mode to an idle mode or a standby mode.
RAID scheduling method comprising a. According to claim 1,
The energy-aware encoding scheduler converts the power modes of the SDD and the HDD from an active mode to an idle mode or a standby mode,
Pipelining the physical data blocks using an online dynamic power management (DPM) method; And
Pipelining the parity blocks using an offline DPM method
Including,
A RAID scheduling method using the online DPM method and the offline DPM method to reduce energy consumption of a distributed storage application. In the RAID (Redundant Array of Inexpensive Disks) scheduling system,
In order to reduce energy consumption of distributed storage applications that individually strip data blocks using SSD and parity blocks using HDD, the power mode of the SDD and the HDD is in an active mode to an idle mode or a standby mode. Energy aware encoding scheduler to convert to
RAID scheduling system comprising a. According to claim 3,
The energy-aware encoding scheduler,
An online dynamic power management (DPM) unit that pipelines the physical data blocks; And
Offline DPM unit for pipelining the parity blocks
Including,
A RAID scheduling system that reduces energy consumption of distributed storage applications by using the online DPM unit and the offline DPM unit. According to claim 3,
Distribution that receives the data chunk from the initiator server, generates the parity block using the CPU at the target storage server, and separately stripes the data block using the SSD and the parity block using the HDD Type storage application
Further comprising, RAID scheduling system.

Images