JP4272275B2 - Storage subsystem - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a storage subsystem having a controller connected to a plurality of hosts or a single host having a plurality of ports, and more particularly, improves reliability and availability in the event of a failure as a storage subsystem. The present invention relates to a technique and a technique for improving the switching performance of processing to an alternate system at the time of failure.
[0002]
[Prior art]
Conventionally, there is a storage subsystem described in Japanese Patent Laid-Open No. 4-215142 as a controller and a storage device such as a disk having redundancy. This storage subsystem is composed of a double system, with one system operating as the active system and the other system operating as the standby system. Then, the storage information of the active disk device is copied to the standby disk device via the shared disk device accessible from both systems, or in the event of a failure of the active system controller, the active system By making it possible to extract the storage information of the disk device, the data integrity at the time of failure of the controller and the disk device has been improved.
[0003]
In addition, a controller that shares a storage medium with multiple controllers and a multi-port configuration can be accessed from multiple paths as shown in the backbone open market. The volume can be accessed uniformly from any controller.
[0004]
[Problems to be solved by the invention]
With the increase in the scale of computer systems, the speeding up of data processing, and the increase in capacity of data, improvement in performance, reliability, and availability of storage control devices is strongly desired.
[0005]
In the prior art, only the reliability and availability that are closed in the control device have been improved by the dual system on the control device side, the dual cache, and the like. Considering the entire computer system, it is essential to duplicate the control device and the host and the access bus to the control device. To realize this, a multiple controller configuration and a multiple port configuration are essential. .
[0006]
However, in order to realize a configuration in which all logical volumes can be uniformly accessed from any controller as in the conventional backbone open market in such a configuration, host data is temporarily stored among a plurality of controllers. It is essential to share cache management information. In order to share cache management information, it is necessary to perform exclusive lock control, management of exclusive information, etc., in order to perform exclusive control by the processor, so in comparison with the performance of one controller configuration, in the multiple controller configuration The rate of performance degradation is large. Therefore, in order to improve reliability and availability, the only alternative was to sacrifice performance.
[0007]
While maintaining a redundant configuration as a system, in order to maintain performance compared with a single configuration that is not redundant, all logical volumes can be accessed from any controller, but lock control and exclusive control between controllers are possible. Etc. are unnecessary. Therefore, as a normal access mode, a method of fixing the responsibility of the logical volume in units of controllers can be considered. However, in the case of simply fixed control, when the host, the host-control device bus, or the control device fails, the responsibility is assigned. It is impossible to substitute processing for a logical volume, and it cannot be said that the configuration is redundant. Therefore, it is necessary that each assigned task is realized with a low overhead during normal times (normal times), and that a shoulder can be substituted when a failure occurs.
[0008]
However, as a connection form to each controller, a single port is connected from a plurality of hosts by, for example, daisy chain connection. Alternatively, the controller has a plurality of port configurations and is connected to each port by a different host. In addition, when each host is configured to be connected to another controller as an alternate path, the controller can be used to switch the logical volume to another controller due to a failure on one host or a bus failure. In the method of switching all the logical volumes in charge of other controllers as a batch, the other hosts cannot perform normal processing access. Therefore, when connecting to a plurality of hosts, switching means suitable for each host and bus failure is required.
[0009]
In recent years, due to the strong needs of fault tolerant, the path switching function, which is a fault tolerant function, is also being supported on the host side. Typical examples include HP's Alternate Link and Safe Path (path switching software). With this function, if the host side has an alternate path connection and recognizes a failure in an arbitrary path, it automatically moves the process to another path (moves the logical volume) and resumes the process. The recognition of a failure is often due to a timeout or the like, and if the processing is delayed for a certain time or more at the first I / O after the processing is transferred to another system, there is a higher risk of being recognized as a failure. Therefore, in the switching process in charge of the controller, I / O is received and recognized. However, it is essential that the subsequent switching process be performed in a short time during which no fault is recognized.
[0010]
[Means for Solving the Problems]
In order to solve the above-mentioned problems, each controller includes a plurality of controllers having a plurality of bus connection ports and a cache mechanism for temporarily storing data, and sharing a storage medium for storing data from the host. Specifying the logical volume (LUN) charge in the control device, specifying the logical volume charge switching method when receiving I / O to the non-responsible controller, specifying the cache allocation method for the logical volume, and A means for designating single writing / double writing to the cache, a processor that controls switching when requested to a logical volume other than the responsible volume, and a control for communicating with another controller, and double writing to the cache / 1 Overwrite switching means to control data transfer, cache memory is divided into controller and LUN units. Means for and use, load information for each LUN, and the LUN of the charge, is realized by the processor to determine the best switching method.
[0011]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0012]
FIG. 1 shows one configuration example of the entire system including the disk array device according to the present embodiment. In FIG. 1, reference numerals 10, 20, 30, and 40 denote host computers, and reference numeral 5000 denotes a disk array device, which is connected to each host by a SCSI bus or the like. 1000 and 2000 are controller units having a dual configuration, 4000 is a disk device group for storing data from the host, and 3000 is an operation panel for designating LUNs. The disk device group 4000 often has an array configuration.
[0013]
Each controller includes four ports 1010, 1020, 1030, 1040, 2010, 2020, 2030, and 2040 that perform bus protocol control, and is connected to each port on the host side. The transfer control units 1100 and 2100 control data transfer between the cache memories 1300 and 2300 and the host, and also control single writing / double writing. The disk array control units 1200 and 2200 control the entire controller A and controller B.
[0014]
FIG. 2 shows the configuration of the cache memory. The cache memories 1300 and 2300 include management information 1400 and 2400, controller A areas 1500 and 2500, and controller B areas 1600 and 2600. Each of 1400 and 2400, 1500 and 2500, 1600 and 2600 is written twice. Since the management information 1400 and 2400 of the cache memories 1300 and 2300 have the same configuration by double writing, the configuration of the management information will be described using the cache memory 1400 on the controller A side as an example. The management information 1400 includes communication information 1410 between controllers, LUN information 1420 for each controller, and data management information 1430 for managing data in the cache. The data management information 1430 is divided into two for each controller.
[0015]
The cache memories 1300 and 2300 are desirably non-volatile for reliability. If the host already has a redundant configuration, for example, if two disk array devices 5000 are prepared and one is used as a current machine and the other as a spare machine, the degree of redundancy on the disk array side If the cache is used as a single write according to an instruction from the user via the operation panel 3000, 1500 and 1600 may be used for the controller A, and 2500 and 2600 may be used for the controller B. it can. In this case, the use efficiency of the cache memory is doubled as compared with the double writing. This single writing and double writing control is realized by the disk array control units 1200 and 2200 setting the transfer modes of the transfer control units 1100 and 2100. The cache memories 1300 and 2300 are used as an RD cache and a WR cache, and can provide high performance. In order to use the WR cache function, it is indispensable to write twice as described above in order to guarantee data in the event of cache memory failure and controller failure.
[0016]
Each host has two ports, each connected to a different controller. In FIG. 1, the port A: 11 of the host 10 is connected to the controller A: 1000 internal port A: 100A via the SCSI bus 100A, the port B: 12-100B-port A: 2010, and so on to another controller. Has been. This connection is a representative example of a configuration having redundancy that can be switched to another system in the event of a host-side failure or bus failure.
[0017]
However, all of the ports of the disk array device 5000 may be connected to different hosts. In this case, there is no redundancy on the host side. When the host has a two-path configuration as shown in FIG. 1, the alternate path may be connected to another port in the same controller. In this configuration, when there is a failure on the controller side, there is no alternate path and access becomes impossible, but when the host side or bus fails, switching to the same controller is possible, switching processing described later, and request by communication Since neither process is required, there is no performance degradation due to switching.
[0018]
Hereinafter, allocation of logical volumes in the configuration of FIG. 1 will be described. Allocation of logical volumes is performed exclusively between controllers. For example, LUNs 0, 1, 2, and 3 are assigned to the controller A1000, and LUNs 4, 5, 6, and 7 are assigned to the controller B2000. The assigned logical volume may be set in advance on the operation panel 3000, or the LUN number of the received command may be assigned after starting up the disk array device. In this way, by dividing the assigned LUN and dividing the cache memory to be used for each controller, exclusive control specific to multi-controller hybrid is not necessary, and high performance can be provided.
[0019]
The plurality of LUNs in the controller may be shared among a plurality of hosts, or may be accessed independently for each host. At this time, the user can specify whether the allocation of the cache memory in the controller is centrally managed in the controller, or divided for each LUN. When LUNs are set independently for each host, if only LUNs under any host occupy cache memory due to access patterns on the host side, capacity differences between hosts, etc., the performance for other LUNs from other hosts Decreases. In order to avoid this phenomenon, the cache memory can be divided in advance for each LUN. This designation can also be set by the operation panel 3000. In addition, when the frequency of access for each LUN changes with time, for example, when accessing LUNs 0 and 1 for daytime operations and LUNs 2 and 3 at night, the LUN unit is used. It is also possible to dynamically change the allocation according to the load state. It is possible to set on the operation panel whether the LUN unit allocation is statically allocated or dynamically allocated. According to the present invention, the performance to a plurality of logical volumes in a multi-host connection environment can provide an equal service to the host in any access pattern from the host.
[0020]
Next, an allocation method for each LUN and a method for changing the allocation method according to the load state will be described with reference to FIG. A minimum management unit for managing use / unuse of the cache memory is called a segment. The data management information manages the number of available segments 1431 for each LUN, the number of currently used segments 1432, the number of unused segments 1433, the actual cache ADR for each segment, and the data ADR from the host. It consists of segment management information 1434. The segment management information has the same number of information as the number of segments. For example, it is assumed that the size of the cache memory 1500 for the controller A is 1,000 segments. Now, it is assumed that four LUNs are allocated to the controller A and each of the controller A is designated to be divided into 250 segments. Therefore, 250 is set as the number of usable segments 1431 for each LUN. When the I / O is received from the host and the cache is used, the segment management information 1434 is set by subtracting from the number of unused segments 1433 in addition to the number of used segments 1432. However, when the cache is used as an RD cache, it is managed as unused, and is managed as being used only when used as a WR cache.
[0021]
Also, the number of I / Os for each LUN received in a certain period is counted. This information may be held in the disk array control unit or in the management information of the cache memory. Based on this information, the load for each LUN is determined, and free segments are migrated from a LUN smaller than the average number of I / Os for each LUN to a larger LUN. That is, it is realized by changing the number of usable segments 1431, changing the number of unused segments 1433, and shifting the location of segment management information. This change process may be performed at regular intervals or when the I / O load from the host is low. With this control, it is possible to dynamically allocate a cache according to the load for each LUN.
[0022]
Next, switching processing at the time of failure will be described. FIG. 3 shows an example of LUN allocation. In this allocation example, not only LUNs are divided for each controller, but LUNs are set independently for each host and each port. With this configuration, when the SCSI bus 100A of the host 10 fails, access to the host-side LUN 0 is performed from the alternate path 100B. At this time, the controller B compares the LUN 0 with the LUN information 1420 for each controller and recognizes that it is not in charge. At this time, when all LUNs for controller A are switched to controller B at once, the processing when the remaining hosts 20, 30, 40 access LUNs 1, 2, 3 from the port A side is also on the controller B side. It will be switched to charge. Therefore, when the controller A receives access to the LUNs 1, 2, and 3 from the hosts 20, 30, and 40, it is determined that the access is to the non-assigned LUN, and the LUN processing is performed on the controller A side that was originally responsible for these LUNs. Processing to switch responsibilities occurs. As described above, the method of switching LUNs under the control of a controller in a batch has a problem that, in such a multi-host connection environment, switching processing frequently occurs when an arbitrary host system fails, and the performance is significantly reduced.
[0023]
In the present invention, in order to solve the above-mentioned problem, it is possible to realize switching of LUN units. When LUN unit switching is performed, when the controller B receives I / O of LUN0 at the time of the above failure, the switching target LUN is set to only the LUN0, so that switching processing is performed even when accessing the controller A from another host. No performance occurs and performance can be maintained.
[0024]
On the other hand, in the above configuration, when the controller A fails, all hosts switch the access right to the port B. Therefore, the number of times of LUN switching is changed by batch switching rather than switching by LUN. Can be reduced, and performance degradation due to LUN processing charge switching processing can be prevented.
[0025]
In addition, when the LUN allocation method shown in FIG. 4 is used, that is, when the LUN is shared between the ports in the controller, that is, between the hosts, the above batch switching is performed at the time of the controller side failure. In the case of a host-side failure or a SCSI bus failure, the switching operation frequently occurs even in the above two switching processes, and the performance is greatly reduced. For example, when the SCSI bus 100A of the host 10 fails, the host 10 issues processing on the controller A side to the controller B using the alternate path 100B. That is, requests for the LUNs 0, 1, 2, and 3 are issued to the controller B.
[0026]
The controller B performs the switching process in response to receipt of the I / O of the non-serving LUN. In the case of batch switching, all LUNs are transferred to the controller B side. Thereafter, when there is a processing request from another host 20, 30, 40 to LUN 0, 1, 2, 3 via the controller A, batch switching occurs again. That is, batch switching occurs every time I / O from the host 10 and I / O from the hosts 20, 30, and 40.
Even in LUN unit switching, exactly the same switching is repeated.
[0027]
In order to solve the above problem, the present invention also proposes a method in which switching is not performed when an I / O for a non-serving LUN is received. This method uses the inter-controller communication information 1410 and requests the partner controller to secure the cache of the partner controller area and read / write data from / to the disk device 4000. When the partner controller completes the request process, the partner controller reports that fact to the inter-controller communication information 1410. The controller itself executes only transfer with the host. According to the present invention, in a configuration in which a LUN in a controller is shared between a plurality of hosts, it is possible to continue processing from a host side or an alternate path at the time of a bus failure without frequent switching operations.
[0028]
In the present invention, the user selects an optimum method from the three modes of the batch switching processing method, the switching processing method for each LUN, and the request processing method to the partner controller without switching in consideration of the access environment of his system configuration. can do. This selection is performed via the operation panel 3000. Further, even when the user selects a switching processing method for each LUN, if a failure occurs in the controller, the LUN processing charge may be switched by the batch switching method.
[0029]
In the present invention, an automatic mode in which the disk array device 5000 side automatically selects from the I / O pattern can be set. A selection method in the automatic mode will be described below.
[0030]
When assigning LUNs to each controller at the time of LUN assignment, that is, not setting for each port, it is impossible to determine whether the LUN in the controller is shared between ports (hosts) or independent. Therefore, the type of access LUN from each port for a certain period is counted, and it is determined whether the share is after a certain period or independent. If it is independent, switching of LUN units is selected, and if it is share, a request method is selected. Since the access pattern may change with time, the above processing is performed after a certain period.
[0031]
Batch switching is effective in a dual configuration with a single port and a single host configuration with an alternate path configuration. In addition, even when a LUN is shared and accessed in a multi-host configuration as shown in FIG. This is effective in a host environment in which other hosts can be controlled to switch to an alternate path at the same time as arbitrary host switching processing.
[0032]
Further, in the storage subsystem of the present invention, when assigning a LUN for processing to each port of a controller, it is possible to assign a plurality of ports in one controller to handle the same LUN. . Also in this case, when the controller receives a processing request for a LUN that is not in charge of processing from the host computer based on the LUN allocation method for a plurality of ports in the controller, the switching method of all LUNs or the LUN A switching method or a request method can be automatically selected.
[0033]
Next, a processing procedure for batch switching processing will be described.
Assume that the controller B receives the non-assigned LUN. At this time, the inter-controller communication information 1410 informs the controller A to that effect. Based on this information, the controller A recognizes batch switching and interrupts the current processing. When interrupted, the fact is notified to the controller B using the inter-controller communication information 1410. When the controller B receives the report from the controller A, the controller B changes the LUN information 1420 for each controller and sets the controller B in charge of all the LUNs under the controller A. Also, the management information for the controller A in the data management information 1430 is all managed on the controller B side. Thus, in the case of batch switching, switching can be performed only by updating management information, and a high-speed switching means can be provided.
[0034]
Next, a switching method for each LUN will be described.
The case where the controller B receives the non-serving LUN and switches from the controller A will be described. Assume that controller B receives a non-assigned LUN. At this time, the inter-controller communication information 1410 informs the controller A to that effect. The controller A recognizes the LUN unit switching based on the information and interrupts the current processing. The interruption ensures that controller A is not currently accessing the cache memory. When interrupted, the fact is notified to the controller B using the inter-controller communication information 1410. When the controller B receives the report from the controller A, the controller B once manages the data management information 1430 for the controllers A and B. The controller B refers to the number of used segments 1432 of the switching target LUN and the management information 1434 of the relevant segment, and uses the number of unused segments for each LUN in the data management information on the controller B side by the number used. From the segment management information, the intra-segment data that is being used by the controller A is copied to an empty segment on the controller B side. The copy process itself can be realized by the transfer control unit 2100 in the controller B.
[0035]
Along with the copy of the data itself, the data management information for each LUN of the switching target LUN on the controller A side is deleted, and the segments originally assigned as the number of usable segments to the LUN are distributed to other LUNs. The distributed segment is registered as the number of unused segments. On the contrary, the data management information on the controller B side adds the switching target LUN information. In the LUN, the number of copied segments is registered as in-use segments, and the number of usable segments is shifted partly from the number of usable segments in other LUNs and registered as free segments. The number of usable segments of other LUNs is also reduced in part, and the number of free segments allocated for copying is also reduced from the number of free segments. When this series of processing is completed, the controller B notifies the controller A of resumption of processing using the inter-controller communication information 1410.
[0036]
In order to realize this control, the number of segments in use in the cache, that is, the number of WR data not reflected in the disk device 4000 is managed by the WR cache function. By giving the highest priority to the writing process to the disk device, it is managed not to increase beyond a certain amount. This control is realized by preferentially processing the write processing when the disk array control units 1200 and 2200 refer to the number of in-use segments for each LUN at a certain period and determine that the number exceeds a certain amount. . With this control, since there are always more than a certain amount of free segments in the cache, it is possible to switch in the LUN unit copy method. This switching process requires a WR data copy process between caches, so that the processing time increases depending on the cache mounting capacity and the WR data amount at that time. However, in recent years, the capacity of the data bus has become very high, and the copy processing between caches generally has a capacity of several hundred MB / S, so this switching processing time is also completed in about several seconds. Therefore, there is no problem in consistency with the host.
[0037]
【The invention's effect】
According to the present invention, in a storage subsystem having a plurality of ports for performing bus protocol control with a host per controller and having a dual controller and cache, a logical volume is independently provided for each controller in connection with a plurality of hosts. Therefore, in any connection form, exclusive control by a plurality of controllers is not necessary in normal conditions, and high performance can be provided compared to a single configuration. Also, if the host connects the alternate path to another path in the same controller in the event of a host-side failure, connection bus failure, or controller-side port failure, switching processing is not required and replacement processing without performance degradation is possible. It becomes possible. In addition, when an alternate path is connected to another controller, since there is redundancy even in the case of a failure on the controller side, switching processing can be performed for all failures. Also, an optimum switching means with a short switching time can be provided depending on the access form from the host to the LUN.
[Brief description of the drawings]
FIG. 1 is a configuration diagram of a control apparatus according to an embodiment of the present invention.
FIG. 2 is a configuration diagram of data in a cache of a controller according to an embodiment of the present invention.
FIG. 3 is an example of allocation of logical volumes between a plurality of hosts according to an embodiment of the present invention, and is an example in which independent logical volumes are allocated for each host and for each port.
FIG. 4 is an example of a method for assigning logical volumes between a plurality of hosts according to an embodiment of the present invention, and is an example in which a logical volume is shared between hosts.
FIG. 5 is a configuration diagram of management information in a cache memory according to an embodiment of the present invention.
[Explanation of symbols]
10, 20, 30, 40: Host computer
11, 21, 31, 41: Port A
12, 22, 32, 42: Port B
100A to 400A: SCSI bus A
100B to 400B: SCSI bus B
1010 to 1040, 2010 to 2040: Ports A to D
1100, 2100: Transfer control unit
1200, 2200: disk array control unit
1300, 2300: Cache memory
1000, 2000: Controller
3000: Operation panel
4000: Disk device group
5000: Disk array device

Claims (8)

A storage subsystem comprising a storage device group having a storage area divided into a plurality of logical volumes, and a plurality of controllers each having a plurality of ports connected to a plurality of host computers,
In each controller, a logical volume in charge of processing in a normal state is preliminarily allocated for each combination of a host computer and a port in accordance with an instruction from a user via an operation panel, and further, the host computer and the storage device A cache memory that temporarily holds data transferred between the groups ;
The cache memory has a storage area for the plurality of controllers,
When a failure occurs in an access path that connects any one host computer of the plurality of host computers and a port of any one controller among the plurality of controllers, the controller removes from the host computer. When a processing request for a logical volume that is not in charge of processing at normal time is received at the time of failure, the controller is in charge of processing of all logical volumes in charge of the controller that was in charge of processing of the logical volume at normal time The processing request contents are communicated to the controller in charge of the logical volume processing or the logical volume processing in a normal state, or the controller that receives the communication processes the logical volume and requests the processing result. Whether to communicate with the original controller Storage subsystem characterized by having an input means or for selecting the treatment method. The storage subsystem of claim 1, comprising:
The storage subsystem, wherein the controller divides a storage area of the cache memory for the plurality of controllers into an area for a logical volume that is in charge of processing by the controller in a normal state. The storage subsystem according to claim 2, comprising:
A storage subsystem comprising input means for designating a capacity of a cache memory area per controller and a capacity of a cache memory area for the logical volume. The storage subsystem of claim 1, comprising:
A storage subsystem for storing data of the plurality of logical volumes in each of cache memories of the plurality of controllers. The storage subsystem of claim 1, comprising:
Input means for selecting whether to store the data of the plurality of logical volumes in each of the cache memories of the plurality of controllers, or to store only the data of the logical volumes in charge of processing when each of the controllers is normal A storage subsystem. The storage subsystem of claim 1, comprising:
When the controller receives a processing request for a logical volume that is not in charge of processing at normal time from the host computer, the storage subsystem is switched by changing only the management information on the cache memory. The storage subsystem of claim 1, comprising:
When the controller receives a processing request from the host computer for a logical volume that is not in charge of processing at the time of failure, it is not stored in the storage area of the storage device group among the logical volumes to be switched on the cache memory A storage subsystem, wherein data is copied to a cache memory in a switching destination controller and a controller in charge of processing of the logical volume is switched. The storage subsystem according to claim 2, comprising:
The storage subsystem according to claim 1, wherein the controller changes a storage area for the logical volume of the cache memory in accordance with an amount of data input to and output from the logical volume.

Images