CN1728665A - Expandable storage system and control method based on objects - Google Patents

Abstract

An extensible object-based storage system and its control method belong to the field of computer information storage technology. The purpose is to overcome the deficiencies in the system structure and user service mode of the existing mass storage system, and propose a method combining file and block methods. Brand new object interface. The system includes I metadata server, N object storage nodes, and M clients connected through the network; three-way communication is realized between clients, object storage nodes, and metadata servers, and the system has strong scalability. Transfer aggregate bandwidth can grow synchronously, and object-based storage enables the system to be adaptive. The invention changes the traditional data management and control mode, the object storage node OSN completes the cumbersome underlying data management function of the traditional file system, and the MS implements metadata management. The system realizes adaptive management function on the basis of object storage.

Description

A scalable object-based storage system and its control method

technical field

The invention belongs to the technical field of computer information storage, and in particular relates to an expandable object-based storage system and a control method thereof.

Background technique

An ideal storage system should have the following five characteristics: security, cross-platform data sharing, high performance, scalability, and intelligence. The three structures of Direct Access Storage (DAS), Network Attached Storage (NAS) and Storage Area Network (SAN) that are widely used today have different degrees of defects, and it is difficult to have the above five characteristics at the same time; the essence of these systems has never changed , that is, the basic transmission unit is block (Block) or file (File), and it lacks the intelligence and other characteristics of the storage system with object (Object) as the interface. As storage requirements increase and storage applications become increasingly complex, block or file-based storage interfaces limit the development of the storage industry.

In many cutting-edge research fields such as earthquake analysis, genetic analysis, and nuclear explosion simulation, the storage capacity is required to be at least 1TB, or even more than 1PB, and the demand for capacity is still growing. At the same time, with the popularization of the network and the promotion of multimedia applications, the requirements for bandwidth are also increasing simultaneously, and some important fields even require the storage system to have a bandwidth of 1TB/S. However, the existing storage system still uses the external storage device mode, which does not meet the needs of network development. The main performance is that the network storage protocol is complex (storage protocol clusters such as SCSI, ATA-5, SATA, FC, iSCSI, iFCP, USB, etc.; network protocol clusters such as IEEE802.x, ATM, TCP/IP, UDP, RTP/RTCP, FTP , GridFTP, SNMP, etc.), complex management, low efficiency, and inconvenience for customers. The existing storage device is a dumb terminal, which only contains data, metadata information is organized and managed by the host, and the storage device only responds passively; the security mechanism is implemented by the host, and the data on the storage device lacks protection (for example, a hard disk is connected to Plugged into another host, its data can still be read); and, as mentioned earlier, the complexity of the storage protocol brings a wide variety of interfaces, and the overhead of protocol conversion is large.

Due to the constraints of the traditional access mode of connecting servers and storage devices, although high-density and large-capacity storage devices can be attached to meet capacity requirements, it cannot simultaneously increase storage bandwidth. Moreover, the I/O data on a large number of storage devices needs to be forwarded to the client through the server. This working mode not only easily leads to peripheral channel congestion, but also increases system Overhead and transmission delays, resulting in lower average data transfer rates and longer wait times for services. Especially when a large number of customers send requests, the above problems will become more prominent, forming a "server bottleneck". Moreover, due to the low intelligence of the storage device and the low adaptive capacity of the system, it cannot actively adapt to changes in the application environment. In this way, the customer's access service cannot be guaranteed, and the burden on the management personnel is increased.

Therefore, it is necessary to construct a mass storage system based on a new object (Object) interface, making full use of the characteristics of object storage, so that the system has good scalability, intelligence, high-speed data transmission rate and server load relief. To meet people's increasingly demanding storage requirements.

Contents of the invention

The present invention proposes an expandable object-based storage system and its control method, aiming at overcoming the deficiencies in the system structure and user service mode of existing mass storage systems, and combining file (file) and block (block) methods to propose A brand-new object interface (Objectinterface); from the perspective of system structure, construct a mass storage system that can meet people's growing demand for capacity and bandwidth, reduce server load, and improve I/O service performance.

An expandable object-based storage system of the present invention includes a metadata server, N object storage nodes and M clients connected through a network;

(1) The metadata server realizes the management of multiple storage objects and the mapping of data on multiple storage objects; each metadata server is connected to the network through the metadata server network interface and the metadata server network channel in turn;

(2) The object storage node is composed of an object-based storage controller connected to J sets of storage devices through J storage device channels, and is responsible for the storage of object data;

A. The object-based storage controller includes a running control module, a storage device interface module and a network interface module, and each module is physically connected through the peripheral component interconnection expansion bus PCI-X;

B. described operation control module comprises CPU, RAM and EPROM, mutually interconnects expansion bus PCI-X physical connection by peripheral component, provides computing power, method operating environment and dispatching ability;

C. The storage device interface module is composed of J storage device interfaces, providing the controller to connect with the storage device;

D. The network interface module includes K network interfaces;

Each object storage node is connected to the network through its own k network interfaces and k network channels;

(3) Each client is connected to the network through the client network interface and the client network channel in turn;

(4) Between the metadata server and the client, the metadata server authenticates and authorizes the client and returns the mapping table of the storage object requested by the client through the network; the client and the storage device of the object storage node perform data transfer through the network Transmission; between the metadata server and the storage device of the object storage node, the metadata management of the storage device by the metadata server is realized through the network; the above-mentioned I, J, K, M, and N are all natural numbers.

The scalable object-based storage system is characterized in that, according to performance and cost, the storage device interface adopts SCSI, FC, IDE or SATA to implement an interface facing the block device level protocol; the storage device is a variety of interfaces The disk drive; the network channel interface can adopt the same or different forms, and is not limited to any network form, each network interface can be connected to the same network or different networks, and accept client requests according to the protocol based on object storage.

The control method of the scalable object-based storage system of the present invention comprises the following steps in sequence:

(1) After the metadata server is initialized, the storage system receives the customer request and performs identity verification on it, and the legal customer request that passes the verification will proceed to the next step, otherwise it will be treated as an illegal customer request;

(2) Determine the type of object requested by the customer and process it accordingly:

Go to step (3) for root object processing requests, go to step (4) for write user object requests, go to step (5) for read user object requests, go to step (6) for partition object requests, go to step (6) for collection object requests (7);

(3) Processing request to the root object:

A. Determine whether the root object requests new registration, if not, perform OSN management of the corresponding object storage node,

B. If a new root object is registered, the object storage node OSN is formatted as an object-based storage logical unit,

C. Create partition objects,

D. The metadata server records relevant information,

E. End of request processing;

(4) Write user object request:

A. The metadata server returns the mapping table and authorization information of the user object to the OSN of the object storage node,

B. The customer establishes a network connection with the relevant object storage node OSN according to the mapping table information and sends an object-based write command,

C. The object storage node OSN responds to the write user object command sent by the client,

D. The object storage node OSN establishes a connection with the metadata server and submits the metadata information changed by processing the user object request,

E. The metadata server updates the relevant metadata information and sends a confirmation message of successful metadata update to the object storage node OSN,

F. End of request processing;

(5) Read user object request:

A. The metadata server returns the mapping table and authorization information of the user object to the OSN of the object storage node,

B. The customer establishes a network connection with the relevant object storage node OSN according to the mapping table information and sends an object-based read command,

C. The object storage node OSN responds to the read user object command sent by the client,

D. The object storage node OSN returns the result generated by reading the user object to the client,

E. End of request processing;

(6) Request for partition objects:

A. Retrieve the partition object,

B. Return the corresponding request information to the customer,

C. End of request processing;

(7) Request for collection objects:

A. Create or retrieve a collection object,

B. Return the corresponding request information to the customer,

C. Request processing ends.

The control method of the storage system of the present invention is shown in FIG. 4 .

The scalable object-based storage system SOSS (Scalable Object-based Storage System) is a solution to provide storage services by storage objects. SOSS divides two parts of the traditional block-based file system—the user-related part and the storage-related part. The user-related part provides the user with an object call interface (managed by the metadata server) through the logical data structure (such as the metadata part of the directory and file); the storage-related part completes the mapping of the data of these directories and files to the logical blocks of the underlying physical device (provided by object storage nodes). In the storage system SOSS, an object is a logical unit of storage, uniquely identified by the global object ID (object ID), and the object contains data (data), attributes (attributes) and methods (methods). The storage system SOSS has four main components: client (Host or Applicationservers), object storage node (Obiect-based Storage Node, OSN) composed of scalable object-based storage controller (Object-Based StorageController, OBSC) , object-based file system (Object-based File System, OFS), metadata server (Metadata Server, MS). Each part is designed as follows:

Client: Applications running on the client see the file system with standard POSIX file system semantics. Special applications can directly access objects without going through the file system.

Object storage node OSN: The object storage node is responsible for storing object data. Object-based storage can load corresponding methods (rules or policies) according to the attributes of the storage objects, so it has intelligent processing capabilities, such as filtering data with object methods before transmitting.

Metadata server MS: realizes the management of multiple storage objects and the mapping of data on multiple storage objects. Among them, the MS implements the management function for the object storage node OSN through the network, and implements the access control function for the client; the data transmission function is implemented between the client and the object storage node OSN.

Object-based file system OFS: Object metadata is used instead of inode data, and the metadata server MS is responsible for the consistency management of object metadata. OFS is based on the standard Linuxext2 file system with a log (joumaling) file system feature.

The storage object (Object) is composed of data (data), attributes (attributes) and methods (methods). The storage system SOSS uses the object property page to describe the characteristics of the object, such as the type of object, the creation time and size of the object, etc.; the method of the object Object properties can be loaded/unloaded according to certain rules (or strategies). Objects are divided into four types: root object (Root Object), partition object (Partition Object), collection object (Collection Object) and user object (UserObject). An object storage node (OSN) has one and only one root object, the partition object contains zero or more user objects (or collection objects), the data area of the partition object only contains the user object ID list, and the attributes of the partition object include The number of user objects in the partition, the space occupied by user objects in the partition, etc. Collection objects are used to quickly retrieve user objects. A partition object can contain zero or more collection objects, and a user object can belong to zero or more collection objects. For partition objects, the status of collection objects and user objects is equal. User objects account for the vast majority of object storage nodes. All objects are identified by partition object ID and user object. When the partition object ID and user object ID values of the object are both zero, it is the root object; when the object partition object ID is non-zero and the user object ID value is zero, it is the partition object. ; When the value of the object's partition object ID and user object ID is not zero, it is a user object or a collection object. An obvious advantage brought by the SOSS storage system is its high intelligence. In the SOSS storage system composed of object storage nodes OSN, the user-related part of the upper layer of the traditional file server is completed by the metadata server MS; while the storage-related part of the lower layer is moved down to the object storage node OSN, and the corresponding device interface is also changed from based on The block or file interface becomes an object-based interface. In this way, the upper layer of the file system is only responsible for mapping logical names such as file names to object IDs, and the load is reduced by about 90%. The mapping between object IDs and disk blocks is completed in the object storage node OSN, and the metadata server MS is not easy to become a processing I/O O requests the "bottleneck" on the path, and can use multiple metadata servers to work in parallel to solve the "bottleneck" problem of the metadata server MS, and the system has strong scalability.

The present invention has following characteristics:

(1) The system has good scalability. Adding a new object storage node OSN only needs to register with the metadata server (the registration process will be described in detail later). Since the metadata server MS only manages metadata, the added load is small, which facilitates the expansion of the mass storage system without affecting system performance, and when the scale of users expands, the aggregate bandwidth of the system increases linearly.

(2) The object storage node can dynamically load the corresponding rules (rules) or policy (policy) according to the change of the application environment, so as to realize such as load balancing and hotspot data migration.

(3) According to the attributes of the storage objects, the storage objects can be organized in multiple object storage nodes so that the operation of the object storage nodes has a high degree of parallelism. In addition to parallel access operations, parallel transmission of different object requests can also be realized.

(4) Realize the management of centralized object metadata and the access of distributed object data to ensure high efficiency of management and high performance of storage.

(5) Data is directly transmitted between object storage nodes and users, which shortens the I/O path, reduces system delay, and improves the average data transmission rate.

Object-based storage can improve the intelligence of object storage nodes through the operation method of loading/unloading objects according to object attributes (such as QoS requirements, etc.). Multiple object storage nodes work together to realize self-organization and self-management of the storage system.

Description of drawings

Fig. 1 is a schematic diagram of the composition and structure of the storage system of the present invention;

FIG. 2 is a schematic structural diagram of an object storage node composed of an object-based storage controller;

FIG. 3 is a schematic diagram of an embodiment of a storage system of the present invention;

FIG. 4 is a schematic diagram of a control method of the storage system of the present invention.

Detailed ways

As shown in Figure 1, the present invention includes I metadata server 200.1～200.i and N scalable object storage nodes 900.1, 900.2, ..., 900.n, metadata server 200.1～200.i through metadata server The network interfaces 221.1-221.i are connected to the network 400, and at the same time are connected to the object storage nodes 900.1, 900.2, ..., 900.n through the network; 121.1~121.k are connected to the network, and M clients 300.1~300.m are connected to the network through client network interfaces 321.1~321.m. The metadata server 200.1～200.i completes the authentication of the metadata server 200.1～200.i to the client through the network channel 220.1～220.i of the metadata server, and the client 300.1～300.m completes the authentication of the client through the network channel 320.1～320.m of the client , authorize and return the mapping table of the storage object requested by the client; the client transmits data through the client network channel 320.1-320.m, and the object storage node transmits data through the respective k network channels 120.1-120.k; the metadata server 200.1-200.i pass through metadata server network channels 220.1-220.i, and object storage nodes implement metadata management of storage systems by metadata servers 200.1-200.i through respective k network channels 120.1-120.k.

FIG. 2 shows a structural block diagram of an object storage node 900 of the present invention. The object storage node 900 of the present invention is composed of an object-based storage controller 100 connected to a storage device 113 through a storage device interface module 112 . The object-based storage controller 100 includes an operation control module 140 , a storage device interface module 112 and a network interface module 122 , and the modules are physically connected through a peripheral component interconnection expansion bus PCI-X130 .

The operation control module 140 specifically includes a CPU 142, a RAM 141 and an EPROM 143, providing computing power, method running environment and scheduling capabilities.

The storage device interface module 112 is composed of storage device interfaces 111.1 to 111.j, providing the connection between the controller and the storage device. The storage device is mainly a disk drive with various interfaces, but is not limited to this type of storage device. The storage device interface adopts SCSI, FC, IDE and SATA according to the performance and cost to implement the interface oriented to the block device level protocol.

The network interface module 122 is composed of network channel interfaces 121.1~121.k. The network interfaces can be in the same or different forms, and are not limited to any network form. Each network interface can be connected to the same network or different networks. The protocol accepts client requests.

The storage system formed by using the storage nodes based on the object interface of the present invention changes the traditional data management and data storage methods. The trivial underlying data operation management in the traditional file system or database system will be moved to the object storage node 900, and part of the data processing originally undertaken by the client application program will also be moved to the object storage node 900 for processing. Using scalable object attribute representation and flexible data organization form, the object storage node 900 can be used to build a storage system suitable for various applications, such as Web service system, database system, file service system, etc.; using the operation method of storing objects And the dynamic scheduling mechanism of the method, it is easy to implement a storage system with self-management functions, such as automatic recovery, automatic load balancing, automatic hotspot migration, etc.

Accompanying drawing 3 is a specific application example of the present invention, what metadata server 200.1 used is IBM xSeries 346Type 8840 server (metadata server can expand according to the needs of actual application), inserts by the Catalyst 3750SERIES of Cisco company by network interface adapter Gigabit Ethernet 400 formed by a mega switch, 4 disks 113.1, 113.2, 113.3, 113.4 of ST173404LC connected to object-based storage controllers 100.1, 100.2, 100.3, 100.4 to form object storage nodes and form an object storage network , the client 300.1, 300.2, 300.3 requests for accessing objects are divided into four types: root object, partition object, collection object and user object. The first two are completed in the metadata server 200.1, and the client's request for the user object includes operations such as reading the user object and writing the user object, firstly through the metadata server to authorize and obtain the mapping table information of the object on the object storage node. Among them, the operation on the user object will involve a large amount of data flow, while other requests are for the management of the object storage system, and the amount of data involved is relatively small. According to the system security authentication mechanism and the control method of object data distribution, when the system is running, the metadata server will check the client identity and authority of each client request, and authorize legal and valid requests.

The storage system processes the client's object-based requests into request processing for root objects, partition objects, collection objects, and user objects. It should be emphasized that before the object storage node OSN can receive and process object-based storage commands, the object storage node OSN needs to register with the metadata server. Accompanying drawing 4 is the control flow of the storage system of the present invention.

Claims (3)

1, a kind of extendible object-based storage system comprises I meta data server, a N object storage nodes and M client of connecting by network;

(1) meta data server is realized the management of a plurality of storage objects, and the mapping of data on a plurality of storage objects; Each meta data server is connected into network by meta data server network interface, meta data server network channel successively;

(2) object storage nodes connects J cover memory device by object-based storage control by J memory device passage and forms the storage of responsible object data;

A. object-based storage control comprises operation control module, storage device interface module and Network Interface Module, passes through periphery component interconnection expansion bus PCI-X physical connection between each module;

B. described operation control module comprises CPU, RAM and EPROM, by periphery component interconnection expansion bus PCI-X physical connection, provides computing capability, method running environment and dispatching each other;

C. the storage device interface module is made of J storage device interface, provides this controller to be connected with memory device;

D. Network Interface Module comprises K network interface;

Each object storage nodes is connected into network by k network interface and k network channel separately;

(3) each client is connected into network by client network interface, client network passage successively;

(4) between meta data server and the client by network finish meta data server to client's authentication, authorize and return the mapping table of the storage object of client requests; Carry out the transmission of data between the memory device of client and object storage nodes by network; Realize the metadata management of meta data server by network between the memory device of meta data server and object storage nodes to memory device;

Above-mentioned I, J, K, M, N are natural number.

2, extendible object-based storage system as claimed in claim 1 is characterized in that, adopts SCSI, FC, IDE or the SATA execution interface towards block device level agreement according to performance, the described storage device interface of cost; Described memory device is the disc driver of each class interface; Described network channel interface can adopt identical or different form, and is not limited to any network form, and each network interface can insert in consolidated network or the heterogeneous networks, accepts client requests by the agreement based on object storage.

3, the control method of extendible object-based storage system as claimed in claim 1 or 2, order comprises the steps:

(1) after the meta data server initialization finished, storage system received client requests, and it is carried out authentication, by the legitimate client request of checking, carried out next step, otherwise handled as illegal client requests;

(2) object type of judgement client requests, carry out respective handling:

Step (3) is changeed in processing request to root object, writes the user object request and changes step (4), reads the user object request and changes step (5), and step (6) is changeed in the request of zone object, and step (7) is changeed in the request of pair set object;

(3) to the processing request of root object:

A. judging whether this root object asks new registration, is not then to carry out corresponding object storage nodes OSN management,

If B. register new root object, then this object storage nodes OSN is formatted as object-based storage logical units,

C. the object that creates the division,

D. meta data server recording-related information,

E. ask processing to finish;

(4) write the user object request:

A. meta data server returns mapping table and the authorization message of the relevant user object of client to object storage nodes OSN,

B. the client sets up network according to mapping table information and related object memory node OSN and is connected and sends object-based write order,

C. object storage nodes OSN customer in response is sent writes the user object order,

D. object storage nodes OSN and meta data server connect and submit the process user object requests to and the metadata information that changes,

E. meta data server upgrades relevant metadata information and sends the affirmation information of metadata updates success to object storage nodes OSN,

F. ask processing to finish;

(5) read the user object request:

A. meta data server returns mapping table and the authorization message of the relevant user object of client to object storage nodes OSN,

B. the client sets up network according to mapping table information and related object memory node OSN and is connected and sends object-based read command,

C. object storage nodes OSN customer in response is sent reads the user object order,

D. object storage nodes OSN will read the result that user object produces and return to client,

E. ask processing to finish;

(6) request of zone object:

A. retrieve zone object,

B. return client's corresponding request information,

C. ask processing to finish;

(7) request of collection object:

A. create or the retrieval collection object,

B. return client's corresponding request information,

C. ask processing to finish.

Images