KR100925334B1 - A method for consuming power reduction in a multimedia server using raid 5 - Google Patents

Abstract

PURPOSE: A method for reducing power consumption in a raid 5 multimedia server is provided to reduce the electric power used in a server by converting a disk in cluster into low power mode. CONSTITUTION: A specific disk enters into a low power mode for each cluster. The data belongs to the same parity group and parity information is pre-fetched. The content of data block of the specific disk which enters into the low power mode is restored through a pre-fetch. A memory size is required for the pre-fetch to convert the specific disk in the low power mode is obtained per each cluster. Within size limit of the memory established, the specific disk entered into the low power mode by cluster order having smaller memory size required in a pre-fetch.

Description

RAID 5 Reducing Power Consumption in Multimedia Servers {A METHOD FOR CONSUMING POWER REDUCTION IN A MULTIMEDIA SERVER USING RAID 5}

The present invention relates to a method for reducing power consumption in a RAID 5 based multimedia server. More specifically, the present invention relates to a method for reducing power consumption by using as many disks as possible in a low power mode by appropriately utilizing available memory and data recovery functions of RAID 5. A method for reducing power consumption in a RAID 5 multimedia server that reduces the amount.

Recently, due to the remarkable development of multimedia technology and network technology, much data service is being performed. In particular, the demand for multimedia data is increasing rapidly, and video-on-demand (VOD) services and user-created-contents (UCC) are good examples. Unlike general data, multimedia data requires high bandwidth and large storage space, so a multimedia server is usually built on a disk array that can consist of tens to hundreds of disks.

Due to the increasing demand for multimedia data, energy consumption of servers has become an important issue. According to a recent U.S. Environmental Protection Agency (EPA) report, data centers consume about 40 times the amount of electricity consumed by typical offices of similar size. In addition, in 2006, the amount of power consumed by servers and data centers in the United States was 61,000,000,000 KW per hour. This figure represents about 1.5 percent of the country's electricity consumption, which is equivalent to 5,800,000 households. This figure is also about twice that of 2000. If that's the case, 2011's electricity consumption will be twice that of 2006. This growing demand for electricity creates serious economic problems for service providers.

Another problem with high power consumption is the generation of heat. It has been reported that the failure rate of a disk drive can double as high as 15 ° C above standby. To address this, cooling systems for high thermal densities are enormously expensive and the cooling system itself adds significantly to the power costs of the data center.

Storage of servers is one of the most energy consuming components. According to recent reports, storage devices consume about 27% of total power. The energy consumed by the disk array also exceeds the energy consumed by the rest of the system, depending on the size of the array. This problem can be exacerbated by the availability of higher speed disks that require more power. In order to reduce power consumption, recent disks have multiple power modes, including several power consumption modes, depending on the state of the disk. In the active mode, platters are spinning and the head is reading or writing data. In the search mode, the head is searching. In idle mode, the disc spins at full speed, but there is no disc requirement. In low power or standby mode, the disk stops spinning completely and consumes much less energy than in any other mode. A commonly used method to reduce energy consumption is to transition the disk to a low power mode after a brief idle state.

Disk power management in mobile devices has been extensively studied, but relatively few attempts have been made to reduce the energy consumption of storage servers. Most existing schemes include a configuration that switches disks to a low power mode whenever possible without impacting performance. These schemes are primarily aimed at increasing the period, during which the disk is in low power mode. However, since the disc must be rotated to return from the low power mode to the active mode, the energy saved by transitioning the disc to the low power mode needs to be greater than the energy required to rotate the disc again. The shortest idle cycle that justifies the cost of rotating energy is called break-even time. However, since video data requires real-time, the data must be read continuously, so the disk does not have a chance to enter the low power mode. In particular, since the length of the video is generally long enough to exceed one hour, the disk can hardly enter the low power mode under the current multimedia server system.

Due to the recent increase in large-capacity VOD systems such as YouTube and mega TV, power reduction is particularly important in VOD systems. The easiest way to allow a disk to enter a low power mode in a multimedia server is to use duplicate data. However, this method has the disadvantage that the available data storage space is only half of the actual data storage space. Therefore, rather than RAID 1, which uses data replication among disk arrays, RAID 5 is used more often, which provides a mechanism to recover disk failures while using disk storage space efficiently. Reduction is an important issue.

The present invention has been proposed to solve the above problems of the conventionally proposed methods, and utilizes the available memory and the data recovery function of RAID 5 appropriately, and puts as much disk as possible into the low power mode to consume the amount of power consumed. It is an object of the present invention to provide a method for reducing power consumption in a RAID 5 multimedia server.

According to an aspect of the present invention for achieving the above object, a method of reducing power consumption in a RAID 5 multimedia server includes entering a specific disk into a low power mode for each cluster, and preloading data and parity information belonging to the same parity group in advance. (Prefetching), and recovering the contents of the data block of the particular disk entering the low power mode.

Preferably, the memory size required for the write-in for switching the specific disk to the low power mode is obtained for each cluster, and the specific disk is arranged in the cluster order of the memory size required for the write-in within the preset memory size limit. It characterized in that to enter the low power mode.

More preferably, all disks belonging to the non-preloadable cluster are operated in the active mode.

More preferably, the memory size required for inserting the specific disk into the low power mode for each cluster may be obtained by using the following equation.

[Equation]

S _k = N _k × BS × (D-1)

Where S _k is the size of the memory for inserting data stored in cluster k, N _k is the number of requests of customers accessing cluster k, BS is the size of the data chunk, and D is the number of disks in the cluster. Respectively.

Preferably, the parity group is a set of associated data chunks and parity chunks for calculating parity, and the size of the parity group is the number of disks provided in the cluster.

More preferably, the parity chunks are distributed and stored in a plurality of disks provided in the cluster.

More preferably, pre-loading data belonging to the parity group, and recovering the required data chunk using the preloaded data chunk and the parity chunk when data chunks stored in the disk entering the low power mode are required. It features.

According to the method for reducing power consumption of the RAID 5 based multimedia server of the present invention, since one specific disk in a cluster can be switched to a low power mode, power consumption of the server can be greatly reduced.

In addition, the method switches a particular disk to a low power mode and recovers it by parity calculation. Since the power consumption required for parity calculation is much smaller than the power that is reduced by entering the disk into a low power mode, power is efficiently supplied. Can be reduced.

In addition, by efficiently using the memory to select the cluster to contain the disk to be switched to the low power mode, thereby improving the power efficiency.

RAID 5 based multimedia server according to an embodiment of the present invention is composed of C clusters, the cluster is composed of D disks. Each file is divided into several blocks and distributed on the disk, and parity information, which is the result of XOR operation of each block, is stored on the disk so that data can be recovered in the event of a disk failure.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a structure showing that the number of disks belonging to each cluster of four (D = 4) in a RAID 5 based multimedia server according to an embodiment of the present invention.

B_i ^k에 의해서 계산되며, 각 패리티 청크들은 디스크에 도 1과 같이 분산 저장한다. 패리티를 계산하기 위해 연관된 데이터 청크와 패리티 청크의 집합을 패리티 그룹 이라고 한다. 예를 들면,B_i ¹, B_i ², B_i ³, P_i ^{1 ,3} 은 동일한 패리티 그룹에 속하며, 패리티 그룹의 크기는 클러스터에 속한 디스크의 개수, 즉 D가 된다.In FIG. 1, assume that video V _i is divided into four disks of cluster 1. V _i consists of sequential chunks, for example B _i ¹ , B _i ² , B _i ³ ,... A parity chunk containing parity information of the jth to kth data chunks is called P _i ^{j, k} . In RAID 5, parity chunks are calculated by performing an XOR operation on data chunks stored on D-1 disks, so parity chunks P _i ^{j, k} are B _i ^j ...

Calculated by B _i ^k , each parity chunk is distributed and stored on disk as shown in FIG. 1. The set of associated data chunks and parity chunks for calculating parity is called a parity group. For example, B ¹ _i, B ² _i, B ³ _i, P _i ^{1, 3} belong to the same parity group, the size of the parity group is the number of the disk, that is, D that belong to the cluster.

The reason for storing parity in RAID 5 is to use parity chunks to recover the failed disk chunks in the event of a disk failure. For example, assume that disk 3 has failed in FIG. 1 and recovers chunk B _i ³ . This can be accomplished by XORing the data stored in the chunks of the remaining disks as follows:

B_i ²

P_i ^{1 ,3} )B _i ³ = (B _i ¹

B _i ²

P _i ^{1 , 3} )

In the conventional method, parity chunks are not read at all (when all disks are running), but are read to recover data chunks of a failed disk when a disk failure is generated. Saving power by entering a low power mode is not very suitable for current multimedia servers because of the long duration of video streams.

For example, suppose that there is only one request for video in FIG. 1, since consecutive chunks corresponding to video (B _i ¹ , B _i ² , B _i ³ ...) must be read continuously from all disks, No disk can enter low power mode.

In the present invention, one particular disk in the cluster is normally switched to the low power mode, and the data chunk stored in the disk that has entered the low power mode is restored using the data chunk and parity chunks stored in the other disk, thereby saving the low power mode of the disk. Make the transition possible. In particular, in the case of multimedia data, successive chunks (eg, B _i ¹ , B _i ² , B _i ³ , ...) must be sequentially read, and according to the present invention, data belonging to one parity group is pre-loaded and low power When data stored in the disk entering the mode is needed, the chunk is restored using the preloaded chunk and the parity chunk.

Memory is required to store the loaded data. Since the total memory size is limited, the data of all clusters cannot be loaded. Therefore, in the present invention, in order to efficiently utilize the memory, the size of the memory required for pre-loading is maintained for each cluster, and the specific disk of the cluster is switched to the low power mode in consideration of a cluster having a small required memory size first. do.

Conventional RAID 5 multimedia servers cannot access a disk in low power mode by accessing all disks and reading data. The proposed method in one embodiment of the present invention enters one disk per cluster into a low power mode and restores a chunk of data stored in the disk entered into the low power mode through a parity operation. To reduce power consumption.

Since the data of the disk entering the low power mode cannot be read, the data chunk or the parity chunk belonging to the same parity group of another disk is utilized to recover the data of the disk entering the low power mode.

FIG. 2 illustrates a structure in which a disk 3 of each cluster is switched to a low power mode in a RAID 5 based multimedia server according to an embodiment of the present invention. Assuming that the multimedia server switches disk 3 to a low power mode as shown in FIG. 2, assuming that the currently required data chunk is B _i ³ , if B _i ¹ , B _i ² , P _i ^{1 , 3} are currently in memory. , B _i ³ can be recovered as follows.

B_i ²

P_i ^1,3 )B _i ³ = (B _i ¹

B _i ²

P _i ^1,3 )

Thus, since the disk 3 does not need to be accessed, it is possible to switch to the low power mode.

In the present invention, all chunks of the parity group belonging to the currently required data chunk are pre-loaded. The multimedia server uses a method of reading consecutive chunks periodically. All chunks of the parity group are preloaded and used to recover data from disks that have entered low power mode.

3 is a diagram illustrating reading data in a RAID 5 based multimedia server according to an embodiment of the present invention.

B_i ² P_i ^{1 , 3}를 활용해서 복구한 뒤 B_i ¹ 청크를 소모한다. B_i ² 청크를 필요로 하는 두 번째 주기 직전에 앞으로 필요할 B_i ⁴ 청크를 선반입하고, B_i ² 청크를 소모한다. 마찬가지 방법으로 계속해서 패리티 그룹 크기만큼의 데이터를 메모리에 미리 저장함으로써, 저전력 모드로 진입한 디스크의 데이터를 복구할 수 있도록 한다.As shown in FIG. 3, a method of reading data in a RAID 5 based multimedia server according to an embodiment of the present invention is a continuous B _i ¹ , which will be needed immediately before a first cycle requiring a B _i ¹ chunk. B _i ² , P _i ^{1 , 3} chunks are pre-loaded, immediately B _i ³ Chunks B _i ¹

B _i ² After restoring using P _i ^{1 , 3} B _i ¹ Consume chunks B _i ² B _i ⁴ , which is needed in the future just before the second cycle requiring chunks Insert chunk, B _i ² Consume chunks In the same way, the data of the parity group size is continuously stored in memory in advance so that the data of the disk which has entered the low power mode can be recovered.

Since there are multiple data requests in a cluster, the total amount of memory required to put a particular disk in the cluster into a low power mode depends on the number of requests from customers accessing the cluster. Let the size of the chunk be BS and the number of requests from customers accessing cluster k to N _k . If the size of memory required for inserting data stored in the cluster k is S _k , the data read from the D-1 disks must be stored in order to recover the data.

S _k = N _k × BS × (D-1)

This, in ascending order according to the value of S _k in the invention, in the memory size limit, input data shelf preferentially less the S _k value clusters and, in the case of the data prefetching non-cluster belongs to the cluster Allow all disks to operate in active mode.

Let's take an example in a RAID 5 based multimedia storage server to verify the effectiveness of this method. Assume that the total number of clusters is 7, the number of disks in a cluster is 5, the server has 1GB of available memory, and the chunk is 1MB. S _k The value is as follows.

When there are 60 requests from customers accessing the first cluster,

S ₁ = 601 MB (5-1) = 240 MB

When there are 80 requests from customers accessing the second cluster,

S ₂ = 801 MB (5-1) = 320 MB

The third cluster has 160 customer requests,

S ₃ = 1601 MB (5-1) = 720 MB

If you have 30 customers requesting access to the fourth cluster,

S ₄ = 301 MB (5-1) = 120 MB

In this way, assume that S _k required by the fifth, sixth, and seventh clusters is S ₅ = 80 MB, S ₆ = 960 MB, and S ₇ = 40 MB, respectively.

First, sorting in ascending order according to the value of S _k , S ₇ , S ₅ , S ₄ , S ₁ , S ₂ , S ₃ , S ₆ . In this case, since the available memory of the server is 1GB, within the allowable range of 1GB, if the number of disks to enter the low power mode according to the present invention is obtained, five disks belonging to the 7,6,4,1,2 cluster will enter the low power mode. Enter and become optimal. If a particular disk in the sixth cluster enters low power mode, only one disk can enter low power mode.

The present invention described above may be variously modified or applied by those skilled in the art, and the scope of the technical idea according to the present invention should be defined by the following claims.

1 is a structural diagram showing that the number of disks (4) belonging to each cluster in a RAID 5 based multimedia server according to an embodiment of the present invention.

2 is a structural diagram showing a disk 3 belonging to each cluster in a RAID 5 based multimedia server according to an embodiment of the present invention converted to a low power mode.

3 is a diagram illustrating reading data in a RAID 5 based multimedia server according to an embodiment of the present invention.

Claims (7)

In a RAID 5 multimedia server having a plurality of clusters consisting of a plurality of disks, Entering a specific disk into a low power mode for each cluster; Prefetching data and parity information belonging to the same parity group to recover contents of data blocks of a specific disk that has entered the low power mode; Method for reducing power consumption in a RAID 5 multimedia server comprising a. The method of claim 1, For each cluster, obtain a memory size required for inserting a specific disk into a low power mode, The method for reducing power consumption in a RAID 5 multimedia server, characterized in that for entering the low-power mode of the particular disk in a cluster order of the memory size required for the preload within a preset memory size limit. The method of claim 2, A method of reducing power consumption in a RAID 5 multimedia server, wherein all disks in a cluster that cannot read information are operated in an active mode. The method of claim 2, The method for reducing power consumption in a RAID 5 multimedia server, characterized in that the memory size required for pre-loading for switching a specific disk into a low power mode for each cluster is calculated using the following equation. [Equation] S _k = N _k × BS × (D-1) Where S _k is the size of the memory to load data stored in cluster k, N _k is the number of requests from customers accessing cluster k, BS is the size of the data chunk, and D is the size of the disk Each number is displayed.) The method of claim 1, The parity group is a set of data chunks and parity chunks associated with each other to calculate parity, and the size of the parity group is the number of disks included in the cluster. The method of claim 5, The parity chunk is distributed and stored in a plurality of disks provided in the cluster power reduction method in a RAID 5 based multimedia server. The method of claim 5, RAID pre-loaded the data belonging to the parity group in advance, and if the data chunk stored in the disk entering the low power mode is needed, the data chunk and the parity chunk is restored by using the preloaded data chunk and RAID. How to reduce power consumption in 5-based multimedia server.

Images