KR100378598B1 - A Buffer Management System and Method for the SAN Environment - Google Patents

Abstract

The present invention is a buffer management technique of a network-connected data storage system in which a plurality of hosts are operated in a cluster form in a network file system. The designed buffer management technique manages individual buffers existing in each host and supports transfer of buffer contents between hosts based on this, enabling hosts to conceptually share each other's buffers. That is, in the existing general Linux system, if a specific data is needed, it checks its buffer and reads it directly from disk if it is not in the buffer, but the buffer management technique of the proposed network-connected data storage system If there is no data in its buffer, it looks at the buffer of the other host and, if there is a host that has the necessary data, receives the data from that host.

In general, transfers over Fiber Channel or Gigabit Ethernet in SAN environments are much faster than disk I / O, so managing global buffers can improve overall cluster performance. Therefore, in the present invention, in order to solve the problem of cluster performance improvement through the global buffer manager module in the existing network file system environment, the global buffer manager and individual buffer manager are newly designed, and the buffer manager and lock manager module are integrated to all the members of the SAN environment. This technique enables sharing of buffer contents on the host.

Description

Buffer management system and method for network-connected data storage system {A Buffer Management System and Method for the SAN Environment}

The present invention relates to a buffer management system and method of a distributed file system connected to a network. More specifically, the present invention relates to a buffer management system and method of a network-connected data storage system in which a plurality of hosts are operated in a cluster form in a network file system. will be.

Due to the proliferation of the Internet, the work environment is changing rapidly, and the amount of data to be stored is exploding. However, existing data storage systems such as client server-type data management systems that connect to one server to store and manage data, or network file systems based on file servers, process exponentially increasing amounts of data. There is a limit to doing so. To solve this problem is a storage area network (SAN). This SAN is a high-speed storage-only network connected by fiber channel. Unlike conventional direct attached storage (DAS) that directly connects a disk to a server, the SAN can be directly connected to a high-speed fiber channel. This provides the advantage of providing high performance, high scalability, high availability and sharing. This advantage of SAN can effectively solve the problem of mass storage device of current data storage system.

In current Linux system, if specific data is needed by user's request or kernel's own request, it checks its buffer and if the required data is not in the buffer, it performs disk I / O and reads data from disk. In general, disk I / O is a time-consuming operation. Therefore, when frequent disk I / O occurs, the system performance decreases accordingly. Therefore, to improve system performance, it is necessary to increase the data hit ratio of the buffer and minimize the number of disk I / Os.

An object of the present invention for solving the problems of the prior art as described above, by combining the individual buffers in each host to configure a global buffer and manage the global buffer list to determine what data is in the buffer of which host The present invention provides a buffer management system and method for a network-connected data storage system that makes it easy to locate a required data on any host.

It is also an object of the present invention to maintain consistency between data when one data exists on multiple hosts at the same time, and to invalidate other data blocks, leaving only the latest version when data changes occur, To provide a buffer management system and method for a network-connected data storage system that supports distributed environments through integration.

Another object of the present invention is to support forwarding of buffers between hosts so that a host in need of data searches for the data in the global buffer as well as in the individual buffer, thereby improving the hit ratio of the data. To provide a buffer management system and method of the storage system.

1 is a structural diagram of a buffer manager in a SAN environment;

2 is a diagram illustrating a buffer manager sharing model between hosts;

3 is a structural diagram of an individual buffer list managed by an individual buffer manager;

4 shows a buffer hash table;

5 is a structural diagram of an individual buffer list and an associated global buffer list;

6 is a structural diagram of a global buffer list managed by a global buffer manager;

7 is an operation flowchart of a buffer transmission and reception process.

In the buffer management system of the network-connected data storage system according to the present invention for achieving the above object, a plurality of hosts are connected through a SAN network to share the data of the buffer connected to each host,

The buffer management system of this networked data storage system,

A plurality of individual buffer managers respectively located in the hosts and managing individual buffer lists for data of a buffer directly connected to each host;

At least one global buffer manager located in a global buffer server selected for integrated management of a plurality of hosts and managing global buffer lists by integrating data of buffers connected to hosts managed by the global buffer server;

An individual lock manager located on each host to manage an individual lock list for locks currently owned by the host, and

Located in the global buffer server includes a global lock manager for managing a global lock list for the data managed by the global buffer server, the global buffer server characterized in that the global buffer manager and the global lock manager integrated management It is done.

In addition, the buffer management method of the network-connected data storage system according to the present invention, the plurality of hosts connected via a SAN network is divided into a separate host and at least one global buffer server, the individual host includes a separate buffer manager and The global buffer server is configured to include a global buffer manager so that each of the hosts share data in the buffer.

The buffer management method in the individual buffer manager of this network-connected data storage system is

A first step of managing an individual buffer list for data of buffers connected in the individual host and registering the buffer with the global buffer manager, and notifying the global buffer manager of changes in the buffer;

Receiving a global buffer list from the global buffer manager and accessing the global buffer manager based on the global buffer list;

When a data block transmission is requested from another host connected to the SAN network, a third step of transmitting the data block to the other host is included.

In addition, the buffer management method in the global buffer manager of the network-connected data storage system according to the present invention,

A first step of managing a global buffer list by collecting the individual buffer lists when an individual buffer list is input from individual buffer managers of the individual hosts;

And providing a global buffer list to the individual hosts.

In addition, the data block retrieval method of a network-attached data storage system, the plurality of hosts connected through a SAN network is divided into individual hosts and at least one global buffer server, the individual host includes an individual buffer manager and the global buffer server Is configured to include a global buffer manager, so that each of the hosts share data in the buffer,

The data block retrieval method on any host of this networked data storage system is

A first step in which the individual buffer manager searches the individual buffer list to find data when a data request is generated from a client;

A second step of providing the client with the found data when the search result data of the first step is found, and requesting data transmission from the global buffer manager if no data is found;

A third step of the global buffer manager searching a global buffer list to inspect a host including the data and providing data providing host information including the data to the data requesting host;

And a fourth step of the data requesting host reading the data from the data providing host and providing the data to the client.

Hereinafter, with reference to the accompanying drawings will be described in more detail "buffer management system and method of the network-connected data storage system" according to an embodiment of the present invention.

1 and 2 are structural diagrams illustrating a buffer manager of a network-connected data storage system according to the present invention.

In the network-connected data storage system, there are a centralized method and a distributed method depending on the type and accuracy of information for global buffer management. The distributed method includes a redundant distributed method and a divided distributed method. In a centralized technique, one server is responsible for managing the global buffer list. The server aggregates the buffer lists of all individual hosts to form a global buffer list and uses them to handle the buffer access requests of individual hosts.

On the other hand, there are several servers for managing global buffers and the servers share the roles to manage global buffers. This distributed technique is divided into redundant distributed and divided distributed techniques according to whether information managed between servers is redundant or mutually exclusive. If the global buffer is managed by the centralized method, the burden is concentrated on the buffer management server, which may cause a bottleneck in performance, and there is a problem in that a large overhead for replacement in the case of a malfunction of the server is large. On the other hand, the distributed technique does not cause bottlenecks because multiple servers divide the overhead of buffer management. The countermeasures against server malfunctions are superior to the duplication management technique, but the disadvantage is that the duplication management requires more overhead than the division management.

Therefore, the network-connected data storage system according to the present invention adopts a split distribution method for buffer management. There are separate buffer managers and multiple global buffer managers, and the global buffer managers manage the buffer lists mutually exclusively. That is, a separate buffer manager (LBM) is installed in each host and manages which data blocks are in the buffers in each host. In addition, it communicates with the global buffer to inform the global buffer manager of the individual buffer list and figure out where the necessary buffers are on the individual hosts.

In addition, some hosts selected as global buffer servers are provided with a global buffer manager (GBM), which identifies the buffer block list of each host for communication with individual buffer managers. Collect and manage global buffer list. The global buffer list managed by each global buffer manager is mutually exclusive. In addition, the global buffer manager (GBM) spreads the global buffer list to the individual buffer manager (LBM) to help each host find the global buffer when it cannot find the required buffer block in the local buffer.

Local Buffer Manager (LBM) performs data retrieval using individual buffer list and global buffer list.

1 is a diagram illustrating the configuration of an individual buffer manager and a global buffer manager. The local buffer manager (LBM) of FIG. 1 extends the existing Linux Linux buffer manager and combines it in a tightly-coupled approach to all hosts where a network-connected data storage system is performed. It is installed and manages which data blocks are in buffers in each host. In addition, it communicates with the global buffer to inform the global buffer manager of the individual buffer list and finds out where the necessary buffers are on the individual host. In addition, the Global Buffer Manager (GBM) exists only on a part of hosts selected as global buffer servers and manages the global buffer list by identifying and collecting the buffer block list of each host for communication with the LBM. It also propagates this list to the LBM so that the global buffer can be found if the necessary blocks cannot be found in the local buffer.

The Local Buffer Manager (LBM) uses three main data structures: That is, it uses individual buffer lists, lists of global buffer servers, and data structures of global buffer lists.

First, the individual buffer manager (LBM) uses a local buffer list that manages the contents of buffer blocks existing in individual hosts. You should always acquire a lock before reading or writing a file to the buffer. That is, each buffer block is also associated with a specific lock. For example, as shown in FIG. 2, if there are six hosts and two global buffer servers and four separate buffers when the file system is initialized, the files on host A and host B Assuming that 1, 2, and 3 exist, each inode ends 1, 3, and 6, and the hash function for the global buffer is Modular 2, the buffer control of file 1 and file 3 is performed in the global buffer server 1. The information is stored and the buffer control information of file 6 is stored in the global buffer server 2. In this way, the global buffer manager divides the load of global buffers according to file inodes. Thus, a lock on a buffer block is eventually associated with an inode or directory entry, which is unique to the file system, and keeping the information consistent is the control information for the lock. In the example, host A has block data for file 1, and at the time of obtaining the block, control information for file 1 exists in the global buffer. Therefore, LBM manages individual buffer lists in association with locks. The actual structure for storing this information is shown in FIG. 3 shows an individual buffer list structure. Each buffer list is a variant of the linked list, has a buffer block, and there is a communication module and a lock module for transmitting control information to the global buffer server. The communication module maximizes the transmission speed by minimizing the control information, and the lock module stores the lock information in each inode. The lock information includes device information, file information, inode information, directory entry information, and the like.

Second, since global buffer lists are managed in a distributed and distributed manner in a network-attached data storage system, each local host must know the list of global buffer servers. Since the global buffer server is the same as the lock server, the buffer server table is the same as the lock server table managed by the lock module. In order to share the roles equally among the servers, we use the hash value for the file inode to determine the server. In other words, if there are n buffer servers, applying a hash function with a value from 0 to (n-1) to each file's inode tells which server manages the lock and global buffer for that file. As an example, the buffer hash table existing in FIG. 4 can specify a server through a hash function when each file's inode is given, and a basic hash function uses a modular function. In addition, it should have weight information so that weight can be assigned according to the performance of the server.

Third, when local hosts acquire a lock on a file, they receive a global buffer list for that file. When using a specific block of data in a file, first look at its buffer and if it does not find it, it looks at the global buffer list that it received when it acquired the lock. If found in the global buffer list, a request is made to send the block to the host. Therefore, the relevant global buffer list is also stored based on the lock. Since a lock can be owned by multiple hosts, the global buffer list for a lock will point to the buffer list of multiple hosts. Figure 5 shows the basic communication header and body part for transmitting a buffer request between hosts when there is a request.

Global Buffer Manager (GBM) uses global buffer list for global buffer management, and it is managed based on lock like individual buffer list. Because global buffer servers and lock servers are the same, there is an array of locks in the global buffer server, and you can see which host owns each lock. At this time, the host that owns the lock maintains a list of buffers.

6 shows a global buffer list managed by a global buffer server. Currently, locks on file F1 are shared by hosts H1 and H2, locks on F2 are shared by H1, H3, and H4, and locks on F3 are owned by H2. In addition, the host H1 owns blocks B1 and B2 of the file F1 and blocks B2 and B3 of the file F2. Host H3 owns blocks B2 and B5 of file F2, and host H4 owns blocks B1 and B5 of file F2. In the example of FIG. 6, when a host requests buffer information for file 2, the global buffer manager sends control information of buffer blocks for host 1 and host 3 hanging on lock (F2) of the global buffer list to the requesting host.

Next, the buffer management method of the individual buffer manager is as follows. The individual buffer manager is responsible for updating individual buffer lists, registering individual buffer lists, managing related global buffer lists, and transferring buffer blocks.

First, the function of updating the individual buffer list is as follows. The individual buffer manager updates the buffer list whenever changes are made to individual hosts. In Linux, the contents of a buffer change when the disk I / O occurs, when the buffer is removed by the swap daemon (bdflush (buffer flush daemon), kupdate (kernel update daemon)). The contents of the buffer can be changed either by receiving a buffer from another host or by releasing the buffer when the lock is released. Therefore, in each of these cases, whenever the contents of the buffer change, the individual buffer list must be updated so that the contents of the individual buffer list and the actual buffer always match.

Next, the function of registering an individual buffer list is as follows. The list of buffers managed in the host must be registered in the global buffer server to have meaning as a global buffer. When a buffer list in an individual host is registered in the global server, the contents of the individual buffer list are reflected in the global buffer list, and other hosts refer to it, and buffer transfer between hosts occurs. To maximize the performance of the global buffer, whenever a change occurs in an individual buffer list, the buffer server must be notified and the buffer server must notify the other hosts. However, this way, messages for buffer list management occur very frequently between hosts, and the buffer server overhead is very large. For example, every time a new buffer block is created or discarded on a host, the host must send a message to the buffer server, which reflects the contents of the message in the global buffer list, and then holds another lock. You must inform all hosts. Thus, many messages occur each time a block in the buffer is changed.

In order to reduce such overhead, the network-connected data storage system according to the present invention registers an individual buffer list in the global buffer server only at a specific time (a time when a lock related message comes and goes). Specifically, if a change is made to the buffer contents of an individual host, the change is reflected only to the individual buffer list, not to the global server. The individual buffer list is continuously reflected to the buffer list, and when it is necessary to send a message to the buffer server (same as the lock server) to acquire or release the lock on the file, the individual buffer that has been managed Send the list appended to the message.

At this time, it is not necessary to send the entire buffer list in the host to the server. Only the list of files managed by the server is registered. For example, suppose you have three lock / buffer servers: s1, s2, and s3. If server s1 manages files f1, f4, server s2 manages f2, f5, and server s3 manages f3, f6, s1 manages only where on the host the blocks of files f1, f4, and s2 manages f2, Manage f5 blocks and s3 manage f3 and f6 blocks. Therefore, if it is necessary to send a lock message from the individual host h1 to the server s1, only the list of the buffer blocks of the files f1 and f4 is registered among the buffer lists managed by the host h1.

Next, let's take a look at the function to manage related global buffer list. Each host constantly updates its own buffer list and informs the server each time it sends a lock message. The server aggregates individual buffer lists to manage the global buffer list and informs individual hosts of this information. Each host manages the global buffer list received from the server and accesses the global buffer based on it.

Next, the function of transmitting the buffer block will be described. When access to the global buffer is required, that is, if the required data block is not in its buffer but in the buffer of another host, each host requests the transfer of the buffer block to the other host. One thing to note is that changes to individual hosts are not immediately propagated to other hosts, so even if you request a buffer transfer based on your global list, you may not actually have the data blocks. This means that the data block has already been discarded in the buffer of the other host. In this case, the host that is requested to transmit the buffer block should inform that there is no corresponding block. On the contrary, if the requested buffer block is found, the block is sent to the requesting host.

Next, the buffer management method of the global buffer manager will be described. The global buffer manager collects individual buffer lists, provides related global buffer lists, and has error recovery and buffer operation scenario functions.

First, let's take a look at the function of collecting individual buffer lists. Each host maintains a separate buffer list, and when it needs to send a lock message to the server, it sends the individual buffer list that it has been managing. The global buffer server receives this list and manages it as a global buffer list. At this time, one global buffer server manages only the list of files in charge. Thus, global buffer lists managed by servers are independent of each other.

Next, we look at the process of providing the relevant global buffer list. In contrast to each host registering an individual buffer list with the server, the server provides the global host with a global buffer list. In this case, as well as the registration of individual buffer lists, it is necessary to decide how often to deliver the global buffer list. If a buffer list is provided to an individual host whenever a change is made to the global buffer list, each host has a more accurate global buffer list, which increases the global buffer hit rate. However, there is an overhead for transmitting the global buffer list. do. Therefore, the network-connected data storage system according to the present invention attaches the global buffer list to the message only when it is necessary to send a lock related message from the server to the host.

Also, when you send a global buffer list from a server to a host, you do not have to send all the global buffer lists managed by the server, but only the buffer lists that the host might use. That is, you only need to send a list of buffers for the files that the host currently has a lock on. For example, suppose server s1 manages files f1, f4, and f7, and host h1 currently holds a lock on files f1, f2. If you want to use file f4 on host h1, you will ask server s1 to lock file f4. Of course, at this time, h1 will send a list of file f1 among the individual buffer lists that it manages to the server s1 with a lock message (file f2 does not need to be sent because server s1 is not responsible). When the server s1 sends the lock permission message for the file f4 to the host h1, the server s1 attaches a list of the files f1 and f4 from the global buffer list. Because host h1 has locks on files f1 and f4, they can access these files.

Next, take a look at the error recovery function. If a particular host or server crashes while the system is running, you need to deal with this anomaly. From a buffer point of view, there are two main types of system anomalies: First, the list of buffer contents of a dead host becomes meaningless because communication with that host is impossible in case of a general host error. Therefore, each global buffer server must remove the buffer list for the dead host from the global buffer list. Also, ordinary hosts should remove the buffer list for dead hosts from their associated global buffer list. Second, in the case of a global buffer server error, the global buffer server also has the function of a normal host. Therefore, even when the global buffer server is dead, the same work must be performed as in a normal system abnormal situation. In addition, you need a way to take over the role of a dead server. As a method of substituting the role of the dead server, there may be a method in which the live servers share the role equally or a method of selecting a new server to play the role of the dead server. In the network-connected data storage system according to the present invention, Choose the latter method. That is, when a global buffer server dies, it randomly selects one of the living hosts (including other global buffer servers) and designates it as a new global buffer server. When a new global buffer server is specified, all hosts change the contents of the buffer server table to reflect that a new server has been allocated. In addition, the newly designated server asks all hosts to reconstruct the global buffer list, gathering the locks that each host is obtaining and the individual buffer lists it manages.

Next, a buffer operation scenario will be described. 7 is an operation flowchart illustrating a buffer transmission and reception process of a network-connected data storage system according to an embodiment of the present invention.

When a buffer request is generated from an arbitrary client (S701), first, request data is found in an individual buffer list (S702), and if found (local hit) (S703), data of the buffer is transmitted to the host (S704). However, if it is not found (S703), the buffer request is sent to the global buffer manager along with the lock request of the lock manager (S705), and the global buffer manager retrieves the request data from the global buffer list and the data exists in the buffer of a host. To find out whether (S706).

If the corresponding data is found in the global buffer list (S707), the global buffer manager sends a buffer request to another host having a buffer including the data (S708), and receives the buffer data from the other host (S709). In step S710, the data is transmitted to the host that made the buffer request. On the other hand, if there is no corresponding data in the global buffer list (S707), a "data not found" message is transmitted to the host that made the buffer request (S712). The individual host retrieves the request data from the individual disk (S712) and receives the request data (S713).

This buffer operation scenario is basically achieved through integration with the lock manager, and the lock manager also interacts with the buffer manager through the global lock manager and the individual lock manager modules. Also, in order to minimize the overhead incurred by setting the global buffer manager, a buffer-related message is added to the lock-related message when the message is exchanged. This somewhat reduces the accuracy and freshness of the global buffer list, but minimizes the overhead.

While the invention has been described above based on the preferred embodiments thereof, these embodiments are intended to illustrate rather than limit the invention. It will be apparent to those skilled in the art that various changes, modifications, or adjustments to the above embodiments can be made without departing from the spirit of the invention. Therefore, the protection scope of the present invention will be limited only by the appended claims, and should be construed as including all such changes, modifications or adjustments.

As described above, according to the present invention, by integrating individual buffers in a SAN environment and managing them as global buffers, the hit ratio of data is improved, thereby improving system performance.

Claims (17)

In the buffer management system of a network-connected data storage system in which a plurality of hosts are connected through a SAN network to share data of a buffer connected to each host, A plurality of individual buffer managers respectively located in the hosts and managing individual buffer lists for data of a buffer directly connected to each host; At least one global buffer manager located in a global buffer server selected for integrated management of a plurality of hosts and managing global buffer lists by integrating data of buffers connected to hosts managed by the global buffer server; An individual lock manager located on each host to manage an individual lock list for locks currently owned by the host, and Located in the global buffer server includes a global lock manager for managing a global lock list for the data managed by the global buffer server, The global buffer server buffer management system of a network-connected data storage system, characterized in that the global buffer manager and the global lock manager integrated management. A plurality of hosts connected through a SAN network are divided into individual hosts and at least one global buffer server, wherein each individual host includes an individual buffer manager and the global buffer server includes a global buffer manager. In the buffer management method in the individual buffer manager of the network-connected data storage system to share the data of the buffer, A first step of managing an individual buffer list for data of buffers connected in the individual host and registering the buffer with the global buffer manager, and notifying the global buffer manager of changes in the buffer; Receiving a global buffer list from the global buffer manager and accessing the global buffer manager based on the global buffer list; And a third step of transmitting the data block to the other host when a data block transmission is requested from another host connected to the SAN network. The method of claim 2, wherein the first step, When a change occurs in the buffer data connected to the individual host, the individual buffer list is updated, and when the lock-related message is transmitted and received with the global buffer manager, the network-connected data storage is added and transmitted. How to manage buffers in your system. The method of claim 3, wherein the first step, If the global buffer manager is 2 or more, the buffer management method of a network-connected data storage system, characterized in that notifying the change only to the global buffer manager for managing the file on which the change occurred. The method of claim 2, wherein the second step, If there is no data block required for the buffer connected in the individual host, the buffer management method of the network-connected data storage system, characterized in that to request the transmission of the data block to another host where the data block exists by searching the global buffer list. The method of claim 2, wherein the third step, And a fourth step of notifying the other host of the absence of the data block when the data block transmission is requested from the other host after the data block is discarded in the buffer. . A plurality of hosts connected through a SAN network are divided into individual hosts and at least one global buffer server, wherein each individual host includes an individual buffer manager and the global buffer server includes a global buffer manager. In the buffer management method in the global buffer manager of the network-connected data storage system to share the data of the buffer, A first step of managing a global buffer list by collecting the individual buffer lists when an individual buffer list is input from individual buffer managers of the individual hosts; And a second step of providing the global buffer list to the individual hosts. 8. The method of claim 7, wherein if the global buffer server is 2 or more, the global buffer lists managed by the global buffer servers are independent of each other. The method of claim 7, wherein the second step, And the global buffer manager transmits the global buffer list in addition to the global buffer list when transmitting the lock related message to the individual host. The method of claim 9, And the global buffer manager transmits only the global buffer list that may be used by the individual host in addition to the lock related message. The method of claim 7, wherein And a third step of removing the individual buffer list of the individual host from the global buffer list when an abnormal situation occurs in the individual host and makes communication impossible. A plurality of hosts connected through a SAN network are divided into individual hosts and at least one global buffer server, wherein each individual host includes an individual buffer manager and the global buffer server includes a global buffer manager. A data block retrieval method in an arbitrary host of a network-connected data storage system for sharing data of a buffer, A first step in which the individual buffer manager searches the individual buffer list to find data when a data request is generated from a client; A second step of providing the client with the found data when the search result data of the first step is found, and requesting data transmission from the global buffer manager if no data is found; A third step of the global buffer manager searching a global buffer list to inspect a host including the data and providing data providing host information including the data to the data requesting host; And a fourth step of the data requesting host reading the data from the data providing host and providing the data to the client. The method of claim 12, If the host including the data is not found in the third step or the data is not present in the data providing host in the fourth step, the data requesting host accesses a disk and reads the data. Data block retrieval method of data storage system. The method of claim 12, wherein the data providing host, And when the data block transmission is requested from the data requesting host after the data block is discarded in the buffer, notifying the data requesting host of the absence of the data block. The method of claim 12, The individual buffer manager sends the lock buffer message with the change of the individual buffer list when the lock related message is transmitted to the global buffer manager, and the global buffer manager sends the lock buffer message with the global buffer list when the lock buffer message is sent to the individual host. Data block retrieval method of a network-connected data storage system, characterized in that. The method of claim 12, And when an abnormal situation occurs in the individual host and the communication is impossible, the individual buffer list of the individual host is removed from the global buffer list. The method of claim 12, If an abnormal situation occurs in the global buffer server and the communication is not possible, a random individual host is designated as a new global buffer server, and the new global buffer server receives the individual buffer list and the individual lock list from the remaining individual hosts, A method for retrieving a data block of a network-connected data storage system, characterized by reconfiguring a buffer list and a global lock list to perform a global buffer function.