Disk subsystem

[0001] The present invention relates to, in a disk subsystem that has one or more hard disks and a disk controller that controls the hard disk and transfers data between the disk controller and a host computer, a technique that distributes and stores data that is input and output from the host computer to the disk controller using the RAID architecture.

[0002] In the disk subsystem that has one or more hard disks and a disk controller that controls the hard disk and transfers data between the disk controller and a host computer, when the data that is input and output from the host computer to the disk controller is distributed and stored using the RAID architecture, particularly the RAID5 architecture, the number of real data items (D) per parity data (P), that is, the number of drives in a magnetic-file stripe can be determined optionally.

[0003] If two hard disks that belong to the same stripe are provided in a fiber loop, however, the two hard disks will not be able to be used at the same time when a fiber loop fault occurred, thereby disabling the recovery of fault data using parity data. Accordingly, to ensure the redundancy of the hard disk when the fiber loop fault occurred, the number of hard disks per fiber loop must be set to 1. Thus, data will be distributed and stored as (n−1) real data items and 1 (one) parity data item for the number of fiber loops (n). A hard disk control unit is formed by collecting m disk adapters that control the fiber loop.

[0004] A disk controller realizes scalability by enabling increased and decreased installation in a unit of this hard disk control unit. When there has 1 hard disk control unit in the disk controller, the number of fiber loops is m, thereby establishing n=m. For example, when the number of disk adapters (m) in the hard disk control unit is 4, the number of fiber loops used is set to 4 and the magnetic-file stripe has the format of 3D+1P. Further, when the number of hard disk control units in the disk controller is also 2 or more, the hard disk control unit is operated for n=m by using the same logic as a single hard disk control unit. By distributing and storing data in the hard disk control unit in this manner, the operation of each hard disk control unit is let to have independence and the increased and decreased installation of the hard disk control unit was enabled without affecting the hard disk control unit in course of system operation.

[0005] Conventionally, since the distribution and storage of data was executed according to the RAID5 architecture using a hard disk control unit, the number of real data items (D) per parity data (P) is determined depending on the number of disk adapters (m) that comprise the hard disk control unit. Accordingly, it was general that a magnetic-file stripe has the format of (m−1)D+1P.

[0006] In recent years, it is requested that while the failure resistance of data is being maintained at the occurrence of a fiber loop fault, the rate of real data in the data stored in the hard disk, that is, a disk utilization rate be increased. In other words, it is requested that k of kD+1P is set to a higher number than (m−1). However, to ensure the failure resistance of data when the fiber loop fault occurred, two or more hard disks cannot be assigned to a fiber loop. Further, if the value of m is increased, the unit price of a hard disk control unit increases. At the same time, to support the format of (m−1)D+1P in the same manner as before, the conventional logic needs to be changed greatly. Accordingly, to suppress the unit price of the hard disk control unit and support a magnetic-file stripe of a conventional format, it is desirable that the number of adapters that comprise the hard disk control unit should be kept set to m.

[0007] Accordingly, a disk subsystem that sets k of kD+1P to a higher number than (m−1) had to be realized using the hard disk control unit of which the number of adapters used as before is m.

[0008] To set k of kD+1P to a higher number than (m−1) with the number of disk adapters per hard disk control unit kept in m without changing it, a parity data item and (jm−1) real data items are created using j hard disk control units and data is distributed and stored in jm hard disks. That is, by using the format of (jm−1)D+1P, a utilization rate of a hard disk is increased maintaining the failure resistance of the hard disk and the conventional format of (m−1)D+1P is also supported. For example, if the number (m) of disk adapters in the hard disk control unit is 4 and the number of hard disk control units (j) used is 2, the data distribution and storage format is set to 7D+1P. In this case, because the number of disk adapters in the hard disk control unit is the same as before, the format of 3D+1P in which data is distributed and stored in a hard disk unit can also be used as before.

[0015] The embodiments of the present invention are described below in detail with reference to the drawings.

[0016] FIG. 1 is an overview of the configuration of a disk subsystem to which the present invention applies. In this embodiment, a disk subsystem 100 has a disk controller 101 and plural hard disks 102, and the disk controller 101 is connected to plural host computers 10. The disk controller 101 has one or more host control unit 103, one or more disk control unit 104, a cache memory 105, and a control memory 106. The host control unit 103 includes one or more processor 107 and one or more host adapter 108, and the disk control unit 104 includes one or more processor 109, one or more disk adapters 110, and a parity generation circuit 111. A host I/F cable 112 connects the host computer 10 with the host adapter 108, and a fiber loop 113 connects the hard disk 102 with a disk adapter 110. A cache access path 114 connects the processors 107, 109 with the cache memory 105, and a control memory access path 115 connects the processors 107, 109 with the control memory 106.

[0017] If a write request is issued from the host computer 10 to the hard disk 102, write data is first transferred to the host control unit 103 via the host adapter 108. The processor 107 writes the write data to the cache memory 105 and writes to the control memory 106 that the write data was written to the cache memory 105. The processor 109 of the disk control unit 104 recognizes that the write data was written to the cache memory 105 by referring to the control memory 106 and the write data is distributed into a plurality of real data. The parity generation circuit 111 generates parity data and writes the plural real data and the parity data to the hard disk 102.

[0018] FIG. 2 is a diagram showing the distribution and storage of data according to the present invention. Two disk control units 200, 210 have 4 processors 201 to 204, 211 to 214, 4 disk adapters 205 to 208, 215 to 218 respectively, and parity generation circuits 209, 219. In the embodiment of the present invention, data is distributed and stored into each of a plurality of hard disks connected to plural disk control units. In this embodiment, a hard disk control unit has 4 disk adapters, and data is distributed and stored using two hard disk control units. Accordingly, the format of a magnetic-file stripe in which data is distributed and stored in a hard disk group 250 is set to 7D+1P (270). Further, as same as the conventional art, the data can be distributed and stored by a hard disk connected to a hard disk control unit according to the magnetic-file stripe having the format of 3D+1P (280). FIG. 2 shows an example of the magnetic-file stripe in which the data is distributed and stored in the format of 7D+1P and the format of 3D+1P. In this example, although parity data is stored in a hard disk connected to the disk adapter 218, for the RAID5 architecture, the parity data is stored in any of the hard disks of the magnetic-file stripe by a predetermined unit.

[0019] FIG. 3 is a block diagram showing detailed data until a processor of the disk control unit writes write data from a cache memory to a hard disk. FIGS. 4 to 6 are flowcharts showing the processing of each processor of the disk control unit that realizes writing of the write data from the cache memory to the hard disk.

[0020] As shown in FIG. 4, the respective processors 201 to 204, 211 to 214 repetitively execute cache data reference processing (401) and disk data reference processing (402).

[0021] FIG. 5 is a flowchart of cache data reference processing. The respective processors 201 to 204, 211 to 214 refer to cache data information 221 on a control memory 220 (501) and monitor whether status 230 is “Unprocessed” or not. When the processor 201 detects the cache data information 221 about which the status 230 is “Unprocessed” (502), the processor 201 changes the status 230 to “Under write data preparation” (503). The processor 201 acquires write data 300 on a cache memory 240 by referring to a cache address 231 and a size 232, distributes the write data 300 into real data 302 to 308, generates parity data 309 using the parity generation circuit 209, and stores the real data 302 to 308 and the parity data 309 in the cache memory 240 (504). Subsequently, the processor 201 calculates a disk address and a size of data storage locations of the respective hard disks 251 to 258 by referring to a device address 233 and the size 232, records eight disk data information 222 to 229 for each hard disks on the control memory 220, and stores the storage addresses of the disk data information 222 to 229 in a disk data information address 234 (505). “Unprocessed” is all recorded in status 235 of the disk data information 222 to 229 to be recorded. The data storage locations of the respective hard disks 251 to 258 are recorded in a disk address 236 and a size 237 and the storage locations of the real data 302 to 308 respectively or the storage location of the parity data 309 are recorded in a cache address. Finally, the processor 201 specifies the status 230 for a “Disk write wait” (506).

[0022] Further, the processors 201 to 204, 211 to 214 monitor whether the status 230 is specified for the “Disk write wait” or not by referring to the cache data information 221 on the control memory 220. In this case, because the status 230 is specified for the “Disk write wait” in the step (506), disk data information is acquired. Because “Unprocessed” is all recorded in the status 235 in the step (505), the cache data reference processing is terminated.

[0023] FIG. 6 is a flowchart of disk data reference processing. The respective processors 201 to 204, 211 to 214 monitor whether the status 235 is “Unprocessed” or not by referring to the disk data information 222 to 229 for the hard disk under control of a self processor on the control memory 220 (601). The processor 202 detects the disk data information 223 for the hard disk 252 in which the status 235 is “Unprocessed” (602) and changes the status 235 to “In course of writing” (603). The processor 202 acquires the real data 303 on a cache by referring to the cache address 238 and the size 237 and transfers real data 303 to the disk address 236 via the disk adapter 206 (604). If transfer is ended normally (605), the status 235 is set to “Already written” (606). The processors 201, 203, 204, 211 to 214 monitor another disk data information 222, 224 to 229 respectively. If the transfer to the hard disks 251, 253 to 258 is ended normally via the disk adapters 205, 207, 208, 215 to 218, the status 235 of the disk data information 222, 224 to 229 is set to “Already written”.

[0024] The respective processors 201 to 204, 211 to 214 re-execute the cache data reference processing. If the status 230 is “Unprocessed” by referring to the cache data information 221 on the control memory 220 (501), the processing 503 to 506 are executed. Subsequently, the respective processors 201 to 204, 211 to 214 monitor whether the status 230 is set to a “Disk write wait” or not. When the processor 203 detects the cache data information 221 about which the status 230 is set to the “Disk write wait” (511), the processor 203 acquires the disk data information 222 to 229 by referring to the disk data information address 234 (512). The status 235 of the data in which the data transfer to a hard disk is ended normally by the disk data reference processing is set to “Already written”. If all of the status 235 are set to “Already written” (513), the status 230 is changed to “Disk already written” (514).

[0025] In the disk subsystem that stores data using the RAID5 architecture, a disk utilization rate can be increased without changing redundancy by enabling an increase in the number of real data items per one parity data while maintaining the compatibility with the conventional method. Further, sequential performance can be improved by increasing the number of hard disks processed concurrently.