KR20040054267A - Distributed and shared RAID system with multiple RAID controllers and management method thereof - Google Patents

Abstract

PURPOSE: A distributed shared RAID(Redundant Array of Inexpensive Disks) system having a multi-RAID controller and a managing method thereof are provided to make hosts share storages through the multi-RAID controller, process an I/O(Input/Output) request transferred from the host in the RAID controller, and consistently manage the database shared by the RAID controllers. CONSTITUTION: The hosts(20-20n) process the request for reading/writing the data transferred from the host. The first optical channel(21) plurally connects the hosts with the RAID controllers(22-22n). An SCI(Scalable Coherence Interface) switch(23) makes the RAID controller exchange a message or the data in order to keep data consistency, and plurally connects respective RAID controller in order to keep the communication between other RAID controllers even if one RAID controller is out of order. The second optical channel(24) switch plurally connects the storages shared by the RAID controllers with the RAID controllers.

Description

Distributed RAID system with multiple RAID controllers and its management method {Distributed and shared RAID system with multiple RAID controllers and management method

The present invention relates to a RAID (Redundant Array of Inexpensive Disks) system, and more particularly to a distributed shared RAID system having a multiple RAID controller and a management method thereof.

Referring to FIG. 1, a conventional RAID system is equipped with two RAID controllers, that is, a RAID main controller 11 and a RAID auxiliary controller 12, in one host 10 and the main controller through a storage area network (SAN). The device 11 and the auxiliary controller 12 are arranged to share multiple storage devices (eg, disks).

The main controller 11 receives or responds to a read or write command transmitted from the host 10, and the auxiliary controller 12 performs fault tolerance in case the main controller 11 has failed. When the tolerant function or the load of the main controller 11 is high, the request of the host 10 is transferred from the main controller 11 to perform a load balancing function.

As described above, the auxiliary controller 12 performing the fault tolerance function stores data transferred after being processed in the main controller 11 in its own memory so that data consistency is maintained between the two controllers 11 and 12. In case of failure of 11, the auxiliary controller 12 serves to keep the raid system in operation.

However, the conventional RAID system in which the auxiliary controller 12 is operated when the main controller 11 fails as described above has the advantage that the reliability of the system is excellent. The disadvantage is that the controller 12 is of no help.

In practice, the auxiliary controller 12 has a certain amount of work performed by the main controller 11 to be higher than a certain degree, so that the read and write requests of the host 10 received by the main controller 11 are forwarded to the auxiliary controller 12. If this is done, the load balance function is performed by informing the master controller 11 of the result of the processing. In this case, the interface with the host 10 is still limited to the master controller 11. Therefore, there is a problem such as maintaining data consistency between the main controller 11 and the auxiliary controller 12.

Accordingly, an object of the present invention is to overcome the above-mentioned conventional problems, and an object of the present invention is to allow multiple hosts to share multiple storage devices through multiple RAID controllers, and to handle input / output requests transmitted from any host. Allows to be handled by any RAID controller, manages data shared by RAID controllers consistently, handles I / O loads from multiple hosts in the RAID controller with the least load among the RAID controllers, and fails the RAID controller. Distributed RAID system with multiple RAID controllers that can recover data stored in internal cache without loss and manage or monitor the system through any of several hosts To provide management The.

The distributed shared RAID system having multiple RAID controllers for achieving the object of the present invention implements a load balancing module that performs a RAID controller load balancing function on a device driver of an operating system for a plurality of storage devices. A plurality of hosts for requesting data reading and writing; A plurality of RAID controllers for processing data read and write requests transmitted from any of the plurality of hosts; A first fiber channel switch connecting the plurality of hosts and the plurality of RAID controllers to be connected to each other in a plurality; The plurality of RAID controllers may exchange messages or data with each other in order to maintain data consistency with each other, and each RAID controller may not disconnect communication between the other RAID controllers even if any one of the plurality of RAID controllers fails. A scalable coherence interface (SCI) switch for connecting the plurality of connections to each other; And a second optical channel switch configured to connect the plurality of storage devices and the plurality of RAID controllers shared by the plurality of RAID controllers to be connected to each other in multiple numbers.

In order to achieve the object of the present invention, a method for managing a distributed shared RAID system having multiple RAID controllers may be performed by examining a load of each RAID controller in advance through a first optical channel switch connected to a plurality of RAID controllers. Transmitting data read and write requests to the plurality of storage devices to a load controller having a low load; Determining, by the RAID controller receiving the read and write request of the host, searching for lock information to determine whether the read and write request of the host is a request to be processed; As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be processed, if the request is a read request to be processed by the host controller, the lock information is transmitted through the second optical channel switch. Reading read data from a storage device accessible by the storage device, storing the read data in a read cache of its local memory, and transferring the read data to a host requesting read; As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be processed, if the request is a write request to be processed by the RAID controller, the write controller receives write data from the write requesting host. Receiving and storing the write cache in its own local memory and transferring the write data to a storage device accessible by the lock information through the second optical channel switch; As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be handled by the host, if the request is not a read request to be processed by the RAID controller, lock information is provided through the SCI switch. Processing the read request obtained by searching and sending the remote read message to the RAID controller, receiving the read data requested by the host from the RAID controller that received the remote read message, and storing it in the read cache of its local memory. Delivering the data to the host requesting read; And a RAID controller receiving the read and write request of the host determines whether the read and write request of the host is a request to be processed by the RAID controller, and if the request is not a write request to be processed by the host, the lock information through the SCI switch. Sends the remote write message to the RAID controller obtained by retrieving the data, receives the write data from the host that requested the write by the RAID controller that received the remote write message, and stores the write data in the write cache of its local memory. To the storage device accessible by the lock information through the second optical channel switch.

1 is a block diagram illustrating a RAID system having a conventional multiple RAID controller;

2 is a distributed shared RAID system having multiple RAID controllers according to the present invention.

Showing a configuration diagram,

3 is a distributed shared RAID system having multiple RAID controllers according to the present invention.

Is an internal block diagram.

<Explanation of symbols for the main parts of the drawings>

10: Host 11: Raid Master

12: RAID auxiliary controller 20-20n: host

21: first optical channel switch 22 to 22n: RAID controller

23: SCI switch 24: second optical channel switch

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

2 and 3, each of the plurality of hosts 20 to 20n implements a load balancing module that performs a RAID controller load balancing function on a device driver of an operating system, respectively, to a plurality of storage devices (eg, disks). Requires reading and writing data.

The load balancing module implemented on the device driver of each of the hosts 20 to 20n is configured by examining the load of each of the RAID controllers 22 to 22n which are multiplexed by the first fiber channel switch 21 below. Delivers data read and write requests for a plurality of storage devices to the low load RAID controllers 22 to 22n, and transfers read and write requests that have not been processed and returned from the failed RAID controllers 22 to 22n. Performs load balancing that passes back to).

The load balancing module is responsible for the number of read or write requests in processing, each of which is a number of factors that determine the load of each RAID controller from each host to select a RAID controller to handle input / output requests from the host to storage. The RAID controller with the least load is selected in consideration of the data size.

The plurality of RAID controllers 22 to 22n respectively handle data read and write requests transmitted from any one of the plurality of hosts 20 to 20n.

The cache management module of each of the RAID controllers 22 to 22n shares a plurality of storage devices and maps a plurality of logical storage devices to the plurality of shared storage devices so that each RAID controller 22 to 22n is identical to each other. Local memory, which consists of a read cache, a write cache and a backup write cache, is managed to handle read and write requests of the plurality of hosts 20 to 20n while sharing data.

The cache management module provides a cache mirroring function to prevent data loss and recover data in case of a failure of the RAID controllers 22 to 22n. The cache management module retrieves cache data, allocates cache data for read cache and write cache, Delete, etc.

The cache management module divides and manages local memory into a read cache, a write cache, and a backup write cache to reduce memory waste for storing mirrored data in a limited local memory, and considers read performance for data already stored in the local memory. Allow multiple RAID controllers 22 to 22n to be stored in the read cache at the same time, but do not provide a mirroring function for the read data and back up only the write cache data, and the backup write cache for each RAID controller 22 to 22n. The RAID controllers 22 to 22n for which data backup relationships are set are defined.

The backup write cache backs up the contents of the write caches of the RAID controllers 22 to 22n for which data backup relationships are established, and is managed in synchronization with the write caches of the corresponding RAID controllers 22 to 22n.

When the size of the read cache and the write cache to be used by the logical storage device is determined, the cache management structure necessary for data management of each cache is allocated to the local memory, and the backup write cache used by the RAID controllers 22 to 22n is considered. The buffer pool used as cache buffer is allocated.

The buffer pool used as the cache buffer is defined as the shared memory shared by the read cache, the write cache, and the backup write cache by allocating the total local memory size, and transferring data by copying the memory between the RAID controllers 22 to 22n. Is performed through the SCI matching device and the SCI switch 23.

Communication between the RAID controllers 22 to 22n is performed by a message passing function by the SCI server module and the SCI client module, which are SCI matching devices.

The SCI server module allocates independent message receiving buffers for each of the RAID controllers 22 to 22n to process messages received from the RAID controllers 22 to 22n that share each other's local memory, and processes the received messages. Create a message receiving process. The message receiving process is responsible for copying the received message and sending the response message to the corresponding RAID controllers 22 to 22n to send the next message, and then searching for the format of the copied message and processing it accordingly. .

The SCI client module is responsible for storing a message in its own message receiving buffer allocated to the corresponding RAID controllers 22 to 22n when it wants to deliver the message to the slave controllers 22 to 22n.

The lock information management module of the RAID controllers 22 to 22n has a plurality of storage devices shared by each of the RAID controllers 22 to 22n. In order to maintain consistency, the data size of the plurality of shared storage devices stored in the local memory is defined in units of cache lines, and a lock indicating which cache of which RAID controllers 22 to 22n is stored and accessible Lock information is allocated for each data of the cache line size.

The lock information management module independently allocates and manages lock information for each logical storage device.

The lock information allocated by the lock information management module of the RAID controllers 22 to 22n is allocated to the data distributed and managed for each RAID controller 22 to 22n, and the lock server is managed while managing the data. It exists only in the RAID controllers 22 to 22n serving as a role, and corresponds to a cache included in the local memory of a specific RAID controller 22 to 22n among the RAID controllers 22 to 22n sharing data for the same logical storage device. If data exists, it is allocated. If the data does not exist in the RAID controllers 22 to 22n, the data is deleted.

The lock server refers to the RAID controllers 22 to 22n that manage data access rights for the allocated data when allocating logical storage devices to be managed for each RAID controller 22 to 22n.

The lock information management module searches for lock information, allocates and deletes lock information, manages a process for requesting lock information, information related to caches of the RAID controllers 22 to 22n in which the data is stored, and a cache of the local memory, and updates the data. The data access inhibit function is performed by the RAID controllers 22 to 22n while the data is in progress.

The lock information management module retrieves lock information prior to processing all read and write requests of the hosts 20-20n, and stores the lock information management structure in a local memory for quick lock information retrieval.

The first optical channel switch 21 is connected so that the plurality of hosts 20 to 20n and the plurality of RAID controllers 22 to 22n are connected to each other multiplely.

The SCI switch 23 allows the plurality of RAID controllers 22 to 22n to exchange messages or data with each other in order to maintain data consistency with each other, and any of the plurality of RAID controllers 22 to 22n has failed. Also, each of the RAID controllers 22 to 22n is connected to each other in multiple connections so that communication between the other RAID controllers is not interrupted.

The second optical channel switch 24 is connected so that the plurality of storage devices and the plurality of RAID controllers 22 to 22n shared by the plurality of RAID controllers 22 to 22n are connected to each other multiplely.

The distributed shared RAID system having a multiple RAID controller according to the present invention configured as described above is managed as shown in FIG. 4 and operates as follows.

First, the plurality of hosts 20 to 20n examine the load of each of the RAID controllers 22 to 22n in advance through the first optical channel switch 21 connected to the plurality of RAID controllers 22 to 22n (S10). The load controller 22-22n transmits data read and write requests for a plurality of storage devices (S20).

At this time, when the read controllers 22 to 22n that have received the request fail, the unread read and write requests from the slave controllers 22 to 22n are returned to the failed controllers 22 to 22n. The host 20 to 20n which has delivered the request transfers the read and write requests back to the other RAID controllers 22 to 22n having the least load among the remaining plurality of RAID controllers 22 to 22n.

As described above, when the read and write requests of the plurality of hosts 20 to 20n are normally transmitted to each of the RAID controllers 22 to 22n, the RAID controller 22 that receives the read and write requests of the specific host 20 to 20n is transmitted. 22n) searches for lock information to determine whether the request for reading or writing the host 20 to 20n is a request to be processed (S30).

In this case, the RAID controllers 22 to 22n that have received the read and write requests of the host 20 to 20n determine whether the request to be read or written by the host 20 to 20n is a request to be processed by the host controller. If the request is a read and write request to be processed by itself, it processes the read and write request (S40).

In fact, if the read request requested by the host 20 to 20n is a read request to be processed by the specific RAID controller 22 to 22n itself, then the RAID controller 22 to 22n may lock information through the second optical channel switch 24. Reads data from a storage device accessible by the storage device, stores it in the read cache of its local memory, passes the read data to the host 20 to 20n that has requested a read, and then stores the read data in the local memory's read cache. The lock information stores that read data requested by -20n) has been stored.

In addition, if the write request requested by the host 20 to 20n is a write request to be processed by the specific RAID controller 22 to 22n itself, the RAID controller 22 to 22n writes from the write request host 20 to 20n. After receiving the data and storing it in the write cache of its own local memory, the write data is transferred to the storage device accessible by the lock information through the second optical channel switch 24.

At this time, the RAID controllers 22 to 22n which have received the write data from the host 20 to 20n and store the write data in the write cache are transferred via the SCI matching device to the current SCI switch 23. After sending the write cache data invalidation message to the other RAID controllers 22 to 22n storing the data and receiving the write cache data invalidation completion message from these other RAID controllers 22 to 22n, backing up the own write data. The data backup message is sent to the RAID controllers 22 to 22n managing the backup data, and the write data is backed up to the backup write cache of the local memory of the RAID controllers 22 to 22n receiving the data backup message.

On the other hand, if the read and write requests requested by the host 20 to 20n are not the read and write requests to be handled by the specific slave controllers 22 to 22n, the slave controllers 22 to 22n are used by other slave controllers 22 to 22n. 22n), the read and write requests of the hosts 20-20n are transmitted to the other RAID controllers 22-22n to process read and write requests of the hosts 20-20n (S50).

In fact, if the read request requested by the host 20 to 20n is not a read request to be handled by the specific RAID controller 22 to 22n itself, the RAID controller 22 to 22n receives lock information through the SCI switch 23. After transmitting the remote read message to the corresponding read request processing RAID controllers 22 to 22n obtained by searching, the read data requested by the host 20 to 20n is received from the slave controllers 22 to 22n that receive the remote read message. It receives the data, stores it in the read cache of its own local memory, and transfers the read data to the host 20 to 20n that has requested the read. At this time, the RAID controllers 22 to 22n receiving the remote read message read the read data from the storage device accessible by the lock information through the second optical channel switch 24 and store the read data in the read cache of the local memory. The read data is transferred to the RAID controllers 22 to 22n that have sent the remote read message, and then stored in the lock information indicating that the read data requested by the host 20 to 20n has been stored in the read cache of its local memory.

In addition, if the write request requested by the host 20 to 20n is not a write request to be processed by the specific RAID controller 22 to 22n itself, the RAID controller 22 to 22n receives lock information through the SCI switch 23. Sends a remote write message to the corresponding write request processing RAID controllers 22 to 22n obtained by receiving the write data from the host 20 to 20n requested by the write controllers 22 to 22n that receive the remote write message. After storing in the write cache of its own local memory, the write data is transferred from the storage device accessible by the lock information through the second optical channel switch 24.

At this time, the slave controllers 22 to 22n that have received the remote write message, via the SCI matching device, via the SCI switch 23, the other slave controllers 22 to 22 that are storing previous data that the current write request process is to be performed. 22n) sends the write cache data invalidation message to receive the write cache data invalidation completion message from these other RAID controllers 22 to 22n, and then to the RAID controllers 22 to 22n that back up and manage their own write data. A data backup message is transmitted to back up its own write data in the backup write cache of the local memory of the RAID controllers 22 to 22n that have received the data backup message.

As described above, a distributed shared RAID system having a multiple RAID controller and a management method thereof according to the present invention allow multiple hosts to share multiple storage devices through multiple RAID controllers, and in particular, a host, a RAID controller, and a storage device. Are independent of each other and can be connected via a fiber channel switch, so it is easy to expand the system by adding RAID controllers or storage devices.

The distributed shared RAID system having a multiple RAID controller and a management method thereof according to the present invention are capable of input / output request to the shared storage device from any host to any RAID controller, and load balancing implemented on a device driver of a host operating system. Through the module, the load of the RAID controllers is reviewed in advance, so that the load controller having the least load controller is selected to handle the input / output request, thereby improving the input / output performance of the system.

The distributed shared RAID system having a multiple RAID controller and a management method thereof according to the present invention are configured to serve as a lock server in which each RAID controller distributes and manages shared data, and the lock server allocates lock information for each cache data. This prevents data corruption by access by multiple RAID controllers to the same data, and saves a copy of the same data in multiple RAID controllers and manages it in lock information so that read requests for the same data can be made. Parallel processing is possible from multiple RAID controllers at the same time in a frequent application environment, and even if a large amount of data processing or data processing for different data is performed simultaneously, multiple RAID controllers process different data for faster data. The advantage is that it can be processed.

The distributed shared RAID system having a multiple RAID controller and a method of managing the same according to the present invention are stored in a write cache of a specific RAID controller managing update information of the cache data in a mirroring cache managed by another RAID controller. Since the same data is stored, data can be prevented from being lost, and even if the specific RAID controller fails, the data can be recovered through another RAID controller storing mirroring data.

What has been described above is only one embodiment for implementing a distributed shared RAID system having a multiple RAID controller and a management method thereof according to the present invention, the present invention is not limited to the above-described embodiment, Without departing from the gist of the present invention claimed in the scope, anyone of ordinary skill in the art will have the technical spirit of the present invention to the extent that various modifications can be made.

Claims (5)

A plurality of hosts implementing a load controller module performing a RAID controller load balancing function on a device driver of an operating system to request reading and writing of data to a plurality of storage devices; A plurality of RAID controllers for processing data read and write requests transmitted from any one of the plurality of hosts; A first optical channel switch connecting the plurality of hosts and the plurality of RAID controllers to be connected to each other in a plurality; The plurality of RAID controllers may exchange messages or data with each other in order to maintain data consistency with each other, and each RAID controller may not disconnect communication between the other RAID controllers even if any one of the plurality of RAID controllers fails. An SCI switch for connecting the plurality of connections to each other; And A second optical channel switch for connecting the plurality of storage devices and the plurality of RAID controllers shared by the plurality of RAID controllers to be connected to each other in a plurality; Distributed sharing RAID system having a multiple RAID controller, characterized in that consisting of. The load balancing module of claim 1, wherein the load balancing module implemented on the device driver of each host is configured to review the load of each RAID controller multiplexed by the first fiber channel switch, and to load the plurality of RAID controllers with the least load. A distributed shared RAID with multiple RAID controllers that delivers data read and write requests for storage devices and load balances the unread and returned read and write requests back to the other RAID controllers. system. The method of claim 1, wherein each RAID controller is Read cache and write to share multiple storage devices and to map multiple logical storage devices to multiple shared storage devices so that each RAID controller can handle the read and write requests of the plurality of hosts while sharing the same data with each other. A cache management module which manages a local memory consisting of a cache and a backup write cache; In order to receive messages transmitted from RAID controllers sharing local memory with each other, an independent message reception buffer is allocated to each RAID controller, and a message reception process for processing the received message is generated. The message receiving process includes: an SCI server module responsible for copying a received message and retrieving the format of the copied message after sending a response to send the next message to the corresponding RAID controller; An SCI client module responsible for storing a message in a message receiving buffer of the slave controller when the message is to be delivered to the slave controller; And Define a data size of a plurality of shared storage devices stored in the local memory in cache line units, and allocate lock information for each cache line size data indicating which cache of which RAID controller the data is stored and accessible. Information management module that maintains the consistency of cache data for each RAID controller sharing the same storage device and accessing the same data. Distributed sharing RAID system having a multiple RAID controller, characterized in that consisting of. A plurality of hosts reviewing the load of each RAID controller in advance through a first optical channel switch connected to the plurality of RAID controllers and transferring data read and write requests for the plurality of storage devices to the least loaded RAID controller; Determining, by the RAID controller receiving the read and write request of the host, searching for lock information to determine whether the read and write request of the host is a request to be processed; As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be processed, if the request is a read request to be processed by the host controller, the lock information is transmitted through the second optical channel switch. Reading read data from a storage device accessible by the storage device, storing the read data in a read cache of its local memory, and transferring the read data to a host requesting read; As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be processed, if the request is a write request to be processed by the RAID controller, the write controller receives write data from the write requesting host. Receiving and storing the write cache in its own local memory and transferring the write data to a storage device accessible by the lock information through the second optical channel switch; As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be handled by the host, if the request is not a read request to be processed by the RAID controller, lock information is provided through the SCI switch. Processing the read request obtained by searching and sending the remote read message to the RAID controller, receiving the read data requested by the host from the RAID controller that received the remote read message, and storing it in the read cache of its local memory. Delivering the data to the host requesting read; And As a result of the RAID controller receiving the read and write request of the host and determining whether the read and write request of the host is a request to be processed by the host, if the request is not a write request to be processed by the RAID controller, lock information is obtained through the SCI switch. Receive the write data from the host that requested the write request by sending the remote write message to the RAID controller by receiving the write request and storing the write data in the write cache of its local memory. Delivering to a storage device accessible by lock information through a second optical channel switch; Distributed sharing RAID system management method having a multiple RAID controller, characterized in that consisting of. The method of claim 4, wherein the plurality of hosts transmit data read and write requests for the plurality of storage devices to the plurality of RAID controllers. After the request is delivered, if the blade controller fails and is not processed from the blade controller, the returned read and write request is returned to another blade controller having the least load among the remaining blade controllers. A method for managing a distributed shared RAID system having a controller.