KR20140047831A - Cloud stroage system using low electric power - Google Patents

Abstract

The present invention relates to a low-power cloud storage system capable of minimizing unnecessary power consumption in the cloud storage system by reducing power consumption by a disk or server to which data access does not occur and rearranging data only when data input/output is requested. According to the present invention, a method for inputting or outputting data to a disk of a data server in the cloud storage system, which includes a metadata server, the data server, a client block, and an administrator block, comprises a step of setting a power mode of the disk; a step of monitoring the set power mode data and actual power mode status of the disk; and a step of inputting or outputting data based on the set power mode data and actual power mode status of the disk. [Reference numerals] (1) Request setting of the power mode of a disk; (110,330,DD) Communication module; (120) Metadata management module; (130) DS management module; (140) MDS management module; (150) Integrated power management module; (160) Metadata database; (2) Monitor the power mode of the disk and manage the setting thereof; (210) Input/output service module; (220) Data storage module; (230) Data management module; (250) Disk power management software module; (3) Provide the power mode status of the disk; (310) Name space processing module; (320) File input/output module; (340) User interface; (4) Data connection; (5) Store data and execute a management command; (6) Controller chip; (7) Buffer; (8) Power connection unit; (9) IDE configuration device; (AA) Administrator; (BB) Administrator interface; (CC) Disk power management module

Description

Low Power Cloud Storage System {CLOUD STROAGE SYSTEM USING LOW ELECTRIC POWER}

The present invention relates to a low-power cloud storage system, specifically, to reduce power consumption of disks and servers that do not have access to data, and to minimize unnecessary waste of power in a cloud storage system by redeploying data when data input / output is requested. It relates to a low power cloud storage system.

Conventional power saving technologies are based on 1) processor power saving technology from the standpoint of single server platform, 2) power conversion efficiency enhancement technology, 3) rack power efficiency enhancement technology, and 4) Research / development.

1) Low power server technology

Low-power server technology is under active research in the field of power-consuming processors and inefficient power converters. Processor power savings technologies can be divided into technologies that reduce power at the operating system level and technologies that reduce processor power at the BIOS level.

A typical technology to reduce power at the operating system level on a microprocessor is Intel's Demand Based Switching (DBS) or SpeedStep, AMD's PowerNow technology. DBS technology is a technology that reduces power by varying the operating voltage and frequency of the process depending on the load. Generally, the load on the processor varies with time in the computer server. Without DBS technology, the CPU consumes a certain amount of power regardless of how much or how little it uses. However, using DBS technology can reduce power consumption when CPU usage is low. DBS and PowerNow technologies require power management in the operating system to save power.

HP's Power Regular technology is a technology that saves processor power in the BIOS stage. BIOS-based power saving technology has the advantage that any platform can operate without the help of a Yoon system or processor, but it is generally less efficient than operating system-level power management technology because it is difficult to efficiently manage the processor state under load.

2) Power conversion efficiency enhancement technology

The biggest power dissipation in the server after the CPU is the power dissipation of the power supply unit (PSU) responsible for AC / DC switching. Generally, the efficiency of the PSU is about 70 ~ 75%, and when the load is less than 40%, the efficiency is lowered. Google operates over hundreds of thousands of computers and pays over $ 2 billion a month for electricity, unifying the voltage generated by the PSU to + 12V, while the voltage regulator module on the motherboard Main boards and PSUs are manufactured to reduce power by 20-30%. Recently, the data center itself is proposing to supply + 12V DC voltage directly to servers.

3) Rack power efficiency enhancement technology

Rack Power reduction technology is a technology that constructs a rack system by effectively estimating power consumption according to the configuration or service of a rack-mounted system. Generally, when configuring the rack, the server is configured considering the maximum power consumption (Nameplate power). However, in general, racks are rarely configured to consume the maximum power consumption. In recent years, a variety of rack configuration policies have been proposed, such as Nameplate approach and Pconfig usage, to save power.

4) Data center power operation technology

In order to reduce power consumption per data center, various measures / researches and suggestions are suggested to improve the power efficiency of the data center along with the power saving technology of the server itself, and a joint research to reduce the power consumption of the data center such as the green grid Projects are also under way. Currently, power efficiency problems in the data center are being addressed as environmental problems, not just the problems of IT service infrastructure. Various technologies are being developed and applied in various fields. To this end, virtualization technologies that increase server utilization, high-efficiency data center power delivery technologies, and high-efficiency data center cooling technologies are expected to become key technologies for future data center power savings.

However, such a power saving technique can reduce the power consumption of the system regardless of the frequency of data use. In general, data is long-term preservation type such as frequently accessed data and infrequently accessed data such as medical data. There is data (data that must be legally stored for a long time), and power is being wasted to maintain data that is not frequently accessed to save power. Accordingly, there is a need for a system that performs power saving based on data usage in a system environment in which multiple storage servers such as cloud storage or distributed network storage systems are configured with a single virtualized image.

The present invention has been made to solve the above problems, and an object of the present invention is to completely cut off the power supply or to provide a minimum power to a node in which data input / output does not occur for a certain period of time among storage nodes of a distributed network storage system. It provides a distributed network storage system that switches to the state of use to minimize the power usage of the entire system.

An object of the present invention is to provide a distributed network storage system capable of activating a node and performing a normal data service by reapplying power when an input / output of data stored in a power-off node is required.

Disclosure of Invention An object of the present invention is to distinguish frequently accessed data from infrequently accessed data, to store frequently accessed data on an active node, and to store infrequently accessed data on an inactive node, thereby minimizing power usage. To provide.

An object of the present invention is to provide a distributed network storage system that can distinguish between frequently accessed data and infrequently accessed data according to the access frequency of data even in an active node and minimize power usage by interworking with the disk power mode. .

An object of the present invention is to provide a distributed network storage system which is divided into an activating node and an inactivating node and normally operating as a single system.

Low power cloud storage system according to the present invention,

A metadata server for storing and managing file metadata;

A data server for processing requests related to data; And

The constituent features include the client block using the storage service.

According to the present invention, in a cloud storage system having a metadata server, a data server, a client block, and an administrator block, a method of inputting and outputting data to and from a disk of a data server,

Setting a power mode of the disk;

Monitoring the set power mode information and the actual power mode state of the disk;

The configuration is performed by the step of inputting and outputting data based on the set power mode information of the disk and the actual power mode.

The cloud storage system having the power control module according to the present invention performs data input / output based on the disk power mode and state of the DS, thereby minimizing the use of server power consumed in driving the disk.

The low-power cloud storage system according to the present invention performs input and output with reference to the current disk power mode and state information in a state in which the disk power mode of the system is preset to minimize power consumption. It can provide faster I / O service.

1 is a block diagram schematically illustrating a general configuration of a distributed network storage system.
2 is a view schematically showing the configuration of a power management system of a cloud storage system according to the present invention.
3 is a block diagram schematically illustrating a configuration of a low power cloud storage system according to the present invention.

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

1 is a block diagram schematically illustrating a general configuration of a distributed network storage system.

In general, a distributed network storage system includes a plurality of Meta Data Servers (MDS) 100, a plurality of DS Servers (Data Storage Servers 200), and a plurality of CLIENT blocks 300 connected by Ethernet to form one large file system. To provide services.

MDS provides node redundancy through HA (High Ability) cluster function to increase service availability. Integrated management node can be configured separately for efficient and integrated management of all system resources and integrated management of resources can be performed from anywhere through Web and CLI (Command Line Interface).

The role of each block is as follows.

MDS ( Meta Data Server , 100)

MDS is a main management system that configures and manages distributed network storage systems. It stores and manages system metadata. When there is a request from the CLIENT block or DS server, it configures the DS server as a virtual single system image to provide a single system image in the CLIENT block. MDS provides functions such as CLIENT block, DS server management and status information collection, volume management, directory storage and management, and metadata operation processing.

MDS is a server that handles file metadata operations of all client blocks. MDS is a user-level daemon process that stores and manages file metadata. It receives and processes requests from the CLIENT block, the DS server. MDS It periodically monitors status information such as network topology and utilization rate, hardware information and storage, CPU information and utilization rate, and provides functions such as file metadata management and operation processing, replication and recovery.

The MDS 100 supporting the single file system includes a communication module 110, a metadata processing module 120, a DS management module 130, an MDS management module 140, and a metadata DB (database) 160. do.

Among them, the communication module 110 receives and confirms a request of a user and a DS connected to the client block 300 through a network, and then makes a processing request to a corresponding module (meta data processing module and DS management module) according to the request type. . When the processing corresponding to the processing request is completed, the result is sent to the client block through the network.

The metadata processing module 120 performs a service for processing metadata such as reading a file set attribute requested by a user connected to the client block 300, searching for an entry, reading an entry attribute, and creating / deleting an entry.

The metadata server 100 supporting a single file system image includes a DS management module 130 that manages all DSs for data storage space, and the DS management module 130 adds and deletes data servers and fails the data server. Detect and manage data server storage space.

The MDS management module 140 manages and terminates the metadata server 100, performs metadata error checking and recovery, and outputs file system statistics information to manage the metadata server.

DS ( Data Server , 200)

DS is a server that handles requests related to file data. As a DS user-level daemon process, file data is stored in chunks with version information, and chunks are created, changed, deleted, replicated from the MDS 100, and access and storage of the data is performed. The request of the client block 300 is received, and the appropriate processing is performed. The DS block periodically notifies the MGT of its status information such as network topology, hardware information and storage, CPU, and network utilization. It also performs functions such as client / MDS input / output services, data storage and management, status information collection and reporting.

The DS 200 includes an input / output service module 210, a data storage module 220, and a data management module 230.

The input / output service module 210 directly processes the file input / output request received from the client 300 without the intervention of the metadata server 1010, thereby reducing the load of the metadata server 100 and reducing the response speed of the client 100. And improves performance. In addition, it receives and processes a request such as file creation / deletion / Truncate from the metadata server 100 or reports the status information of the data server 200 to the metadata server 100, wherein the chunk validity such as a version mismatch check is performed. The test is also performed.

The data storage module 220 stores the version information together with the data of the file to provide a function of determining whether the chunk has an old invalid version and whether the recording of the chunk is normally completed. By calculating the version information or the CRC value of the chunk at the request of the client 300 or the metadata server 100, it helps to check the validity of the chunk.

The data management module 230 performs a chunk duplication operation to another data server 200 according to an instruction from the metadata server 100, and reports the error code to the metadata server 100 as a result.

The client 300,

A client is a block that uses storage services. File system APIs and file system utilities provide file services for the system. That is, the client 300 performs a function of providing a file system API, setting and releasing a file system mount, processing a namespace, and inputting / outputting a file.

The client 300 includes a namespace processing module 310, a file input / output module 320, a communication module 330, and a user interface.

Since the namespace processing module 310 stores and manages all file metadata of the file system in the UCS FS system in the metadata server 100, the file or directory creation, deletion, retrieval, renaming, etc. It performs the function of processing namespace operation through metadata server block. It has file system mount / unmount function, namespace processing function, and platform independent operating environment function.

The file input / output module 320 performs a function of writing file data to the data server block 100 or reading file data from the data server block 100. In detail, the file input / output module and the online data server block extension are performed. And a removal support function, a fault tolerance function for the failure of the data server block 100, and a file sharing semantic function.

The communication module 330 performs socket communication processing between the metadata server 100 block and the data server 200 block. In particular, to efficiently communicate with the metadata server 100 block, a predetermined number of socket connection pools are created, and communication processing with the metadata server 100 block is performed using the predetermined number of socket connection pools. The communication with the data server 200 is a form of continuously transmitting or receiving data after one read or write request in order to optimize sequential access to the file data.

One or more storage nodes (data store, 240) configured in the DS 200 support the Intelligent Platform Management Interface (IPMI) protocol as in the configuration of a general purpose server, and focus on various hardware resources around the baseboard management controller (BMC). Configured to perform monitoring and control. The power management software communicates with the BMC and the hard disk controller to collect and manage power related information, and performs power control of the system through communication with the central main management server.

2 is a view schematically showing the configuration of a power management system of a cloud storage system according to the present invention.

The low-power cloud storage system according to the present invention can separately configure an integrated power management module to efficiently manage all system resources and perform integrated management of resources anywhere through the web and the command line interface (CLI). .

The DS (data storage) node supports the Intelligent Platform Management Interface (IPMI) protocol as in the configuration of a general-purpose server, and is configured to monitor and control various hardware resources around the baseboard management controller (BMC).

BMC is a microcontroller mounted on the metadata server's motherboard to monitor and manage the hardware resources of the server based on the Intelligent Platform Management Interface (IPMI) protocol. It provides an interface and collects information such as temperature, fan speed, power status and CPU status from sensors supporting various types of IPMI protocols installed in cloud storage systems. Also, the configuration of the hardware resources can be changed and managed through communication with another hardware controller supporting IPMI. The BMC can send alarms from the sensor to the system administrator through the network, and can perform system management such as setting change, rebooting, etc. through communication with the remote system administrator.

The power management software 250 of the DS server communicates with the BMC and the disk controller to collect and manage power related information, and performs power control of the system through communication with the integrated power management module 150 of the MDS.

Cloud storage system having a power module according to the present invention can apply the disk power mode to the disk without a long-term input and output and relocate the data can reduce the power for data management, maintenance.

3 is a block diagram schematically illustrating a configuration of a low power cloud storage system according to the present invention.

 The low power cloud storage system according to the present invention allocates reads and writes to disks based on the power mode state of the disks of each DS. That is, the power consumption of the DS is minimized by first selecting a currently operating disk or a disk which consumes little power to newly drive and performing data read / write.

Low power cloud storage system according to the present invention is a metadata server (MDS, 100). Data server (DS) 200 and client block 300.

The metadata server 100 may include a communication module 110, a metadata management module 120, a data server management module 130, a metadata server management module 140, an integrated power management module 150, and a metadata database ( 160).

The data server 200 includes an input / output service module 210, a data storage module 220, a data management module 230, a plurality of disks 240, and a disk power management SW module 250.

The data server 200 further includes a Diabis driver and an adapter.

The client block 300 includes a namespace processing module 310, a file input / output module 320, a communication module 330, and a user interface 340.

The low power cloud storage system according to the present invention further includes a manager block 400, which includes a disk power management module 410, a communication module 420, and a user interface 430.

The disk 240 used in the cloud storage system according to the present invention can set the power mode of the disk in order to minimize the power use, and this function is called an APM (Advanced Power Management) function. The configuration change and monitoring of the disk power mode can be performed through the disk standard interface ATA (AT Attachment).

The power mode and the power used of the disk are as follows.

Online) Online (random read / write): The state of general use, the state where the disk and the head are set to operate continuously, the state that consumes the most power.

Ii) Normal Idle: The head is turned off after a few seconds after reading and writing operation. It saves about 40% of the power compared to the online state.

Un) Unload Idle: If the hard disk 240 does not work for a few minutes, the head is moved to the lamp and some circuit power is cut off. It consumes about 50% less power than the online mode. If a work request comes in, the disk returns to Online mode, which requires about 1 second to return to Online mode.

RPM) Low RPM Idle: If the hard disk 240 is not used for a longer time, it goes into an additional step to save more electricity, i.e. lower the spindle motor speed (RPM) after moving the head to the ramp, usually online Lower the spindle motor speed by about 50% of the mode. This saves about 20-30% more power than Unload mode, but it takes about 5 seconds to return to Online mode.

Drive) Hard Drive Standby: It is a mode that cuts off the power of the entire hard disk 240. It consumes 25% more power than Low RPM mode, but it takes about 10 seconds to return to Online mode.

Cloud storage system according to the invention, the disk power mode setting step; Disk power mode setting information and status monitoring step; I / O of data is managed by inputting / outputting data based on the disk power mode.

In order to manage the data input and output of a plurality of disks of the DS, first set the power mode of each disk, when setting the disk power mode of the DS node through the manager interface 430, the disk of the DS (200) receiving this The power management SW module 250 sets the power mode for the corresponding disk 240.

Next, the disk power mode setting information and status are monitored. The DS power management SW module detects the power mode setting information and the actual disk power mode state of each corresponding disk, and stores the corresponding information. 150, the integrated power management module of the MDS periodically stores the information in the metadata DB.

Next, in order to input and output data based on the disk power mode, when an input / output request for data is transmitted from the client 300 to the MDS 100, the metadata management module 120 of the MDS 100 may include a metadata DB ( Based on the disk power mode setting information stored in the 160 and the actual disk power mode state, the DS and the disk 240 to determine the data input / output are to be performed and the information about the determined DS and the disk 240 is determined by the client 300. To pass). The client 300 receiving the information on the determined DS and disk 240 communicates with the DS 200 to perform input / output for the determined DS and disk.

At this time, there are three methods for determining the DS and the disk 240 to perform data input / output based on the disk power mode setting information and the actual disk power mode state.

First, a method of setting the priority of the DS and the disk 240 to perform data input / output based on the preset disk power mode, and the predetermined power mode of each disk is Online, Normal Idle, Unload Idle, and Low RPM. One of Idle and Hard Drive Standby.It has priority of data input / output in the order of Online, Normal Idle, Unload Idle, Low RPM Idle, Hard Drive Standby.

Secondly, the priority of the DS and the disk 240 to perform data input / output based on the current disk power mode state is set. The current power mode of each disk is Online, Normal Idle, Unload Idle, Low RPM Idle. It is one of Hard Drive Standby and has priority of data input / output in order of Online, Normal Idle, Unload Idle, Low RPM Idle, Hard Drive Standby.

Third, a method of setting priorities of the DS and the disk 240 to perform data input / output based on both a preset disk power mode and a current disk power mode state.

As described above, in the data input / output method of the cloud storage system having the control module according to the present invention, when the manager block requests the disk power management module to set the disk power mode, the disk power management module may execute the power mode of each disk. Monitoring and configuration management, and transmit to the integrated power management module, the metadata management module determines the data placement policy based on the power mode status of the disk to the namespace processing module, DS stores and manages data from the client block Receives commands to perform data storage and management.

Since the low power cloud storage system according to the present invention performs data input / output based on the disk power mode and state of the DS, it is possible to minimize the server power consumption consumed for driving the disk, and the disk power mode of the system is pre-set to minimize the power consumption. I / O is performed by referring to the current disk power mode and status information in the set state. Therefore, the power consumed by the disk can be substantially reduced and faster I / O service can be provided than when I / O is performed without using the information. have.

The foregoing description is merely illustrative of the technical idea of the present invention and various changes and modifications may be made without departing from the essential characteristics of the present invention by those skilled in the art. Therefore, the embodiments disclosed in the present invention are not intended to limit the scope of the present invention but to limit the scope of the present invention. The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

Claims (13)

A metadata server for storing and managing file metadata;
A data server for processing requests related to data; And
Containing client blocks that use storage services
Low power cloud storage system.
The method according to claim 1,
The metadata server includes:
A communication module for receiving a request of a client block and a DS and transmitting a processing result;
A metadata management module for processing a metadata service;
A data server management module managing a plurality of data servers;
A metadata server management module that performs MDS startup and shutdown, MDS error checking and recovery, and file system statistical information output;
Metadata database; And
An integrated power management module for periodically storing the set power mode information and the actual power mode state in a metadata database
Low power cloud storage system.
The method according to claim 1,
The data server
An input / output service module that creates, deletes, and truncates a file according to a file input / output request of an MDS;
A data storage module for storing data and version information of the file;
A data management module for performing chunk replication to another DS at the request of an MDS;
A plurality of disks each capable of setting a power mode; And
A disk power management SW module that configures and monitors disk power mode and provides disk power mode status to the MDS.
Low power cloud storage system.
The method according to claim 1,
The data server further includes a device driver and an adapter
Low power cloud storage system.
The method according to claim 1,
The client block includes a namespace processing module, a file input / output module, a communication module, and a user interface.
Low power cloud storage system.
The method according to claim 1,
And a manager block having a disk power management module, a communication module and a user interface.
Low power cloud storage system.
In a cloud storage system including a metadata server, a data server, a client block, and a manager block, a method of inputting and outputting data to and from a disk of a data server may include:
Setting a power mode of the disk;
Monitoring the set power mode information and the actual power mode state of the disk;
Is performed by inputting / outputting data based on the set power mode information of the disk and the actual power mode.
Input / output method of cloud storage system having a control module.
The method of claim 7, wherein
Setting the power mode of the disk,
Setting a power mode of a disk of the DS node in the manager block through the manager interface and transmitting the power mode to the data server; And
The disk power management SW module of the data server includes setting a power mode for the disk.
Input / output method of cloud storage system having a control module.
The method of claim 7, wherein
Monitoring the set power mode information and the actual power mode state of the disk,
Transmitting, by the disk power management SW module of the DS, the set power mode information and the actual power mode state of each corresponding disk to the integrated power management module of the MDS;
The MDS's integrated power management module storing the set power mode information and the actual power mode state of each corresponding disk in a metadata DB;
Input / output method of cloud storage system having a control module.
The method of claim 7, wherein
The step of inputting and outputting data based on the set power mode information and the actual power mode of the disk,
Transmitting an input / output request for data from the client to the MDS;
Determining, by the metadata management module of the MDS, a DS and a disk to perform data input / output based on the disk power mode setting information and the actual disk power mode state stored in the metadata DB;
Delivering information about the DS and the disk determined to perform data input / output to the client;
Client performing I / O to the determined DS and disk.
Input / output method of cloud storage system having a control module.
11. The method of claim 10,
Determining a DS and a disk to perform data input and output,
Determining the DS and disk to perform data I / O first based on the preset disk power mode or disk's current power mode status or the preset disk power mode and disk's current power mode status.
Input / output method of cloud storage system having a control module.
The method of claim 7, wherein
The power mode of the disk is one of online, normal idle, unload idle, low rpm idle, and hard drive standby modes.
Input / output method of cloud storage system having a control module.
The method of claim 7, wherein
Monitoring the set power mode information and the actual power mode status of the disk is performed through the disk standard interface ATA (AT Attachment)
Input / output method of cloud storage system having a control module.

Images