|Publication number||USRE36846 E|
|Application number||US 08/583,773|
|Publication date||Aug 29, 2000|
|Filing date||Jan 11, 1996|
|Priority date||Jun 18, 1991|
|Also published as||CA2066154A1, CA2066154C, EP0519670A2, EP0519670A3, US5278838|
|Publication number||08583773, 583773, US RE36846 E, US RE36846E, US-E-RE36846, USRE36846 E, USRE36846E|
|Inventors||Spencer W. Ng, David W. Palmer, Richard S. Thompson|
|Original Assignee||International Business Machines Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (10), Non-Patent Citations (2), Referenced by (49), Classifications (9), Legal Events (4)|
|External Links: USPTO, USPTO Assignment, Espacenet|
The present invention relates to redundant arrays of disk drives, particularly to recovery from degraded redundancy by rebuilding data of error-affected tracks causing the degradation into spare tracks or disks.
Patterson et al in the article "A CASE FOR REDUNDANT ARRAYS OF INEXPENSIVE DISKS (RAID)", ACM 1988, Mar. 1988, describe several arrangements for using a plurality of data-storing disk drives. Various modes of operation are described; in one mode the data storage is divided among the several drives to effect a storage redundancy. Data to be stored is partially stored in a predetermined number of the disk drives in the array, at least one of the disk drives storing error detecting redundancies. For example, four of the disk drives may store data while a fifth disk drive may store parity based upon data stored in the four disk drives. Such a redundant array of disk drives may provide high data availability by introducing error detecting redundancy data in one of the disk drives. For example, four data blocks (one data block in each of the four drives) are used to compute an error detecting redundancy, such as a parity value; the computed error detecting redundancy is stored as a fifth block on the fifth drive. All blocks have the same number of data bytes and may be (not a requirement) stored in the five disk drives at the same relative track locations. The five drives form a parity group of drives. If any one of the drives in the parity group fails, in whole or in part, the data from the failing drive can be reconstructed using known error correcting techniques. It is desired to efficiently rebuild and replace the data from the failing disk drive while continuing accessing the drives in the array for data processing operations.
The disk drives in a parity group of drives may act in unison as a single logical disk drive. Such a logical drive has logical cylinders and tracks consisting of like-located cylinders and tracks in the parity group drives. In such array usage, the data being stored is partially stored in each of the data-storing drives in an interleaved manner in a so-called striped mode. Alternately, the disk drives and their data in the parity group may be independently addressable and used in a so-called independent mode.
Whenever one of the disk drives in a single-parity array fails, even though data can be successfully recovered, the fault tolerance to error conditions is lost. To return to a desired fault tolerant state, the failing disk drive should be replaced or repaired and the affected data content rebuilt to the desired redundancy. It is desired to provide control means and methods for effecting such rebuilding of data and its redundancy to remove the error from a partially or wholly failed disk drive in a parity array of disk drives.
It is an object of the invention to complete rebuild of data in a parity group of disk drives to a fault tolerant state after detecting loss or degradation of the fault tolerant state by a partially or wholly failed disk drive in a parity array of disk drives.
It is another object of the invention to complete a rebuild of data to a fault tolerant state in a relatively non-intrusive manner while accesses to a parity array continue for data storage and retrieval.
In accordance with the invention, failures in a redundant array of disk drives is remedied by rebuilding the error-affected data using any one of a plurality of methods and apparatus any of which enable a continuing use of the disk drive array for information handling and data processing. Such rebuilding may use any or all of the methods and apparatus. A first method and apparatus is a variable rate rebuild which schedules rebuilds at a rate in a detected inverse ratio to a current or pending rate of disk drive usage or accessing within a parity group. Upon completing each scheduled rebuild, this method and apparatus also preferably takes advantage of any idle time of the array by continuing rebuild if there is no waiting access. A second method and apparatus effects rebuild during predetermined array idle times by starting a non-scheduled rebuild of a predetermined portion of the error-affected data. A third or opportunistic method and apparatus detects a need for a data rebuild during a usual access to the array. All three methods and apparatus are preferably used in conjunction with each other.
The above-described methods and apparatus rebuild data onto a scratch or new disk drive which replaces a disk drive in error. A purpose of the rebuild is to restore redundancy of the array. These methods and apparatus also apply to a partially failed disk drive in which the error-affected data are rebuilt in a different track or zone of the disk drive in error; in the latter rebuild, data in the non-failing disk drives may also be moved to corresponding zones or tracks in the respective drives.
The foregoing and other objects, features, and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings.
FIG. 1 illustrates in simplified form an information handling system employing the present invention.
FIG. 2 is a graph illustrating the principles o variable rate rebuilding in the FIG. 1 illustrated array of disk drives.
FIG. 3 is a machine operations chart showing detecting errors in the FIG. 1 illustrated array and priming the system for rebuilding.
FIG. 4 is a machine operations chart showing activating any one of three rebuild methods or apparatus.
FIG. 5 is a simplified machine operations chart showing selection of a rebuild method and apparatus.
FIG. 6 is a diagrammatic showing of a disk recording surface as may be used in practicing the present invention.
FIG. 7 is a diagrammatic showing of a data structure usable in practicing the present invention.
FIG. 8 is a diagrammatic showing of a bit map control for effecting rebuild.
FIG. 9 is a machine operations chart showing rebuild using a variable rate method and apparatus.
FIG. 10 is a machine operations chart showing rebuild using an array idle time method and apparatus.
FIG. 11 is a machine operations chart showing rebuild using the opportunistic rebuild method and apparatus.
Referring now more particularly to the appended drawing, like numerals indicate like parts and structural features in the various figures. Host processor(s) 10 (FIG. 1) are respectively connected to one or more controller(s) 11 by host to peripheral interconnection 12. A plurality of parity arrays 13, 14 and 15 are connected to controller 11 by a usual controller to peripheral device connection 17. Each of the FIG. 1 illustrated arrays 13-15 include five disk drives 20-24, no limitation thereto intended. Four of the disk drives 20-23 store like-sized blocks of data of one data unit. The block sizes may vary from data unit to data unit. A data unit can be an amalgamation of files, one file, graphic data, and the like. A fifth disk drive 24 is a parity or error detection redundancy storing drive P. The redundancy is a parity data block having the same size as the corresponding data blocks of the data unit. The redundancy is computed based upon any algorithm, including simple parity for example, using the data in the data blocks stored in drives 20-23. All data blocks of any data unit in disk drives 20-24 may be stored in the same relative tracks in all of the disk drives; i.e. all data blocks may be stored in track 16 of all the drives, for example; while this storage arrangement simplifies data management, it is not essential for practicing the present invention.
Disk drives 20-24 form a parity group, with disk 24 being dedicated to storing parity blocks. This general arrangement is known as RAID 3 and RAID 4 architecture, see Patterson et al, supra. Alternately, the storage of the parity block scan be rotated among all disk drives in the parity group with no single drive being designated as the parity drive. This latter storage arrangement is known as RAID 5 architecture. The present invention is equally applicable to any of these architectures, plus other architectures.
When storing data to any one or more of the disk drives 20-23, a new parity value is computed and stored in disk drive P 24. For efficiency purposes, it is desired to simultaneously record data in all four disk drives 20-23, compute parity and record parity on disk drive P 24 as the data are being stored on the data drives; in a rotated parity arrangement, parity data are stored in the appropriate disk drive.
Host processors 10 and controller 11 both participate in the above-described machine operations; it is to be understood that the combination of host processors 10 and controller 11 comprises a computer means in which programming resides and is actuated for effecting the practice of the present invention in the illustrated embodiment; such programming is represented by the machine operation charts of the various figures. Such programming can be a separate part of the FIG. 1 illustrated system or can be embodied in ROM, loadable software modules, and the like.
Rebuilding data in a parity group on a disk drive that replaces a failed disk drive, either a spare or true replacement is achieved using a combination of methods, first scheduled rebuilds are controlled by a variable rebuild rate method, an idle time rebuild occurs during any idle time of the parity array and an opportunistic rebuild method is invoked upon each access to a replacement drive for the failed drive for accessing a non-built data storing area. This description assumes that a failed drive (a drive having a number of non-recordable data storing tracks/clusters of sectors, a failed mechanical part that prevents accessing and the like is a failed drive) has been replace with a scratch drive or disk using known disk drive and disk replacement procedures.)
Before proceeding with a detailed description of the methods, the principles of the variable rate rebuild method is described with respect to FIG. 2. When this method is active, rebuild disk accesses (input/output operations or I/O's) are commanded or scheduled at a rate which varies inversely to the current level of I/O activity. FIG. 2 illustrates how the rate of rebuild scheduling is ascertained. Such rebuilding is interleaved with host processor 10 or other disk drive accesses, as will become apparent. A desired response time T is determined for the parity group to be managed. Such a response time is determined using known system analysis techniques, or the rate can be arbitrary and capricious. The five curves 30, also respectively labelled 1-5, show the variation of average response time (vertical ordinate) with the total I/O rate represented on the horizontal ordinate. The total I/O rate is determined by the activity of the host processors 10. The I/O rate is repeatedly monitored in predetermined constant measurement periods. The measured I/O rate determines the rebuild rate for the next ensuing measurement period. The measured rate during each measurement period is the computed average I/O rate for the measurement period of the parity group. When the I/O rate is higher than R1, then no rebuilds are scheduled during the next ensuing measurement period. During such a measurement period rebuilds may occur using either the idle or the opportunistic rebuild methods. The rebuild schedule rate for one measurement period is next listed using the FIG. 2 chart as a guide. An I/O rate between R1 and R2, one rebuild is scheduled; upon measuring an I/O rate between R3 and R2, two rebuilds are scheduled; a measured I/O rate between R4 and R3 results in three rebuilds to be scheduled; a measured I/O rate between R5 and R4 results in four rebuilds being scheduled while lower I/O rates than R5 result in five rebuilds being scheduled. The maximum number of scheduled rebuilds is five; any number can be used as the maximum. In the illustrated embodiment, a minimum size rebuild is one track. The information represented by the FIG. 2 chart is maintained for the parity array in the computer means for effecting scheduled rebuilds.
FIG. 3 illustrates reading data from one of the parity arrays 13-15, detecting rebuild criteria and priming the FIG. 1 illustrated system for rebuilding data. A usual read operation occurs at machine step 35 as initiated by other machine operations. At machine step 36 controller 11 (disk drives may contain error detecting facilities as well as the controller or host processors) detects errors in the data read from any of the disk drives 20-23; such errors are attempted to be corrected in a usual manner. At machine decision step 37, controller 11 determines whether or not the error corrections by the error redundancies in the individual disk drives 20-23 were successful and whether or not fault tolerance was degraded even with a successful error correction. If error corrections were successful (high quality redundancy may still be indicated for some purposes and degraded redundancy may be indicated for other purposes, as will become apparent), then, assuming fault tolerant redundancy is not degraded for requiring a rebuild (NO degradation detected in machine step 37), machine operations proceed to other operations; no rebuild activity is indicated. On the other hand, if any one of the disk drives did not yield correctable data errors, which include a failure to respond, fault tolerance degradation is indicated. With the parity disk P 24, such data errors can still be corrected by reading the parity redundancy of the block from disk drive P 24, then computing the correct data from the data successfully read from the other drives and the parity redundancy. To achieve this known parity correction, the parity block stored in disk drive P 24 is read into controller 11. Then the data are corrected in machine step 39 using the known parity correction procedures. Such correction can occur in either a host processor 10 or in controller 11. At this point in time, the redundancy for the data unit being read has been removed. Then, at machine step 40, the parity correcting unit (host processor 10/controller 11) determines whether or not the parity correction is successful. Whenever the parity correction is unsuccessful, a subsystem error is flagged in a usual manner. Then, recovery procedures beyond the present description are required. Whenever the parity correction is successful, then at machine step 41, if it is determined that there is insufficient degradation of the fault tolerance effecting redundancy, other machine operations are performed; if it is determined that fault tolerance is unacceptable (a disk has failed, for example), then a rebuild is indicated.
The present invention enables maintaining the desired redundancy without undue interference with day-to-day operations. FIG. 4 illustrates the concepts of the present embodiment of the invention. The general arrangement of FIG. 4 can be thought of as establishing an interrupt driven examination of rebuild needs in a system. Machine step 45 represents monitoring and indicating I/O (input output) rate of operations for each parity group 13-15 of disk drives. At predetermined times, as will become apparent from FIG. 9, from such rate monitoring and indicating, a rebuild need is detected at machine step 46. Such detection may merely be a global rebuild flag or any entry in any of the FIG. 8 illustrated bit maps. If a rebuild is needed, then at machine step 47 a later described valuable rate rebuild is scheduled. If a rebuild is not needed, then other machine operations are performed.
Similarly, machine step 50 represents monitoring for idle time for idle time in any one of the parity groups 13-15. If idle time is detected, such as no pending access requests nor free standing operations are being performed, then machine step 51 represents detecting a rebuild need. When a rebuild need is detected, then at machine step 52 a later-described idle time rebuild is effected. If no rebuild is required, other machine operations ensue.
Likewise, machine step 55 represents monitoring for a failed access, read operation or write operation in any one of the parity groups 13-15 or any access to a known failed drive. Upon detecting such an error, a rebuild need may be indicated as described for FIG. 3. Then at machine step 56 the rebuild needs are detected. On one hand, if the parity correction described in FIG. 3 was successful, a rebuild may be deferred, then from machine step 56 other operations ensue. If a rebuild is required, then the later-described opportunistic rebuild operations of machine step 57 are performed.
It is to be appreciated that the FIG. 4 illustration is tutorial; actual practical embodiments may differ in substantial details without departing from the principles of the present invention. Interleaving the plural rebuild techniques can follow several variations. The determination of when and the total extent of a rebuild may substantially affect a given design.
FIG. 5 shows one method of selecting between two of the three illustrated data rebuild techniques. The selection procedure is entered at path 60 from other machine operations based upon any one of a plurality of criteria, such as a time out, time of day, number of accesses, the later-described rebuild schedule of the variable rate rebuild, whether or not bit map of FIG. 8 indicates any rebuild need and the like. Such a selection could typically reside in a dispatcher or other supervisory program (not shown). At machine decision or branching step 61, the type of rebuild needed to be evaluated is selected. Machine step 61 represents a program loop function controlled by software counter 62. Entry of the procedure at path 60 resets counter 62 to a reference state, counter 62 enables the decision step 61 to first evaluate an idle rebuild at machine step 65 as detailed in FIG. 10. If none of the parity arrays 13-15 are idle or there is no need for any rebuild (bit maps of FIG. 8 are all zeros), then operations return to machine step 61, counter 62 is incremented to a next value. This next value causes decision step 61 to effect evaluation of a variable rate rebuild at machine step 66 as detailed later in FIG. 9. The rebuild scanning may return to FIG. 5 from FIG. 9 to reexecute machine step 61 and increment counter 62. Other rebuild procedures may be employed (not described) as represented by numeral 67. Again, upon completing the rebuild evaluation, machine operations returning to the FIG. 5 procedure results in another incrementation of counter 62 and execution of machine step 61. Since the program loop scanning of the procedures has been completed, other machine operations are performed as indicated by numeral 68. The order of scanning the rebuild procedures or methods is arbitrary. As shown in FIG. 11, the opportunistic rebuild procedure is always entered from a disk accessing operation. Any method of scanning rebuild procedures may be employed for selecting any one of a plurality of rebuild procedures.
FIG. 6 is a diagrammatic plan view of a disk in any of the disk devices 20-24; a plurality of such disks are usually stacked to be coaxially co-rotating in the respective devices. All tracks on the respective disks having the same radial location constitute a cylinder of such tracks. When employing a traditional fixed block architecture, each disk 70 may be envisioned as a plurality of disk sector-indicating radially-extending machine-sensible lines 71. Each disk sector between the lines is addressable as a data-storing unit. In a count-key-data (CKD) disk a single radially-extending track index line is used. Each disk has a multiplicity of addressable circular tracks, or circumvolutions of a spiral track, reside on each disk 70. A track 72 may be error affected requiring a partial rebuild of the array. In disk 70 the data contents of the error affected track 72 may be reassigned to track 73; in a rebuild the data contents of all tracks 72 in the respective disk devices 20-24 are similarly reassigned to their respective tracks 73. In one mode, the data contents of a cylinder of tracks in which track 72 may be reassigned to a cylinder of tracks including track 73. In another mode, only the contents of a single track are reassigned. When a disk device is totally replaced, then the data from all of the remaining devices 20-24 are used to compute the data for the replaced disk. The decision when to replace a disk device that is partially operable may be based upon the number of unusable or bad tracks on the device, the character of the error causing failure, and the like.
FIG. 7 illustrates a data structure for one implementation of the variable rate rebuild method. The objective of this method is to maintain during the rebuild period at least a minimum level of subsystem performance, i.e. response time to received I/O requests. Three registers or data storage locations 80-82, either in a host processor 10 or controller 11, store control information need to effect the variable rate rebuild method in the respective parity arrays 13-15. Each register 80-82 is identically constructed, the criteria information may be different to accommodate arrays having different performance characteristics or system usages. Field rate 83 stores a number indicating the rate of rebuild, i.e. one rebuild per second, two per second, etc. Field AVG-IO 84 stores the average I/O response time, possibly expressed in terms of its corresponding I/O request rate, in a predetermined measuring period. The I/O response time or request rate is used to compute the rebuild rate. Fields 85-88 respectively store rebuild rates for the I/O request rates T-1 through T-4 for various rebuild rates, no limitation thereto intended. The total number of disk accesses to a parity array is an indication of response; the greater the number of access requests, the lower the desired rebuild rate. Thresholds T-1 through T-4 correspond to decreasing numbers of access requests rates and indicate higher and higher rebuild rates. Threshold T-1 indicates an access rate greater than which would result in no rebuild being permitted by the variable rate rebuild method. Threshold T-2 indicates an access rate greater than T-3 and smaller than T-1 and which permits one rebuild access (i.e. rebuild one track of data) during a constant request rate measuring period. Similarly, threshold T-3 indicates an access rate greater than T-4 and smaller than T-2 and which permits two rebuild accesses during the constant rebuild rate measuring period. As request rates continue to decrease, corresponding increases in rebuild rates occur. A predetermined maximum rebuild rate for the system may be established. In another implementation of the variable rate rebuild method, an average response time can be directly measured during each successive measuring period. If the measured response time is slower than a desired response time, the rebuild rate to be used during the next successive measuring period is reduced. If the measured response time is shorter than the desired response time, the rebuild rate used in the next successive measuring period is increased. Alternately, if I/O access queues exist, then the rebuild rate may be selected to be inversely proportional to the length of the access queues for the respective parity arrays. Any of the above-described measurement techniques may be employed for establishing the rebuild rate control information stored in fields 83 and 84. Numeral 89 indicates that additional criteria may be used for determining a rebuild rate which is inversely proportional to accessing/free-standing array operations.
Which tracks needing rebuild are maintained in bit maps 95-97 (FIG. 8) respectively for parity arrays 13-15. The rows 105, 106, 107 ... of bit-containing squares 99 respectively indicate sets of logical tracks on one recording surface of disk 70 (FIG. 6). The columns 100, 101, 102 ... of bit-containing squares 99 respectively represent logical cylinders of the logical tracks. Each logical track includes one physical track in each device 20-24 and each logical cylinder includes one physical cylinder in each device 20-24. When any one of the parity groups 13-15 is providing complete redundancy, then all of the squares 99 in the respective bit map 95-97 contains binary 0's. Any track needing a rebuild, whether as part of a complete rebuild of a disk device or a partial rebuild, is indicated by a binary 1 in the square or bit position 99 of the respective bit map. Scanning the bit maps for ascertaining rebuild needs follows known techniques. An index or written-to-bit-map value (not shown) may be used to indicate that a respective bit map either contains at least a binary 1, all binary 0's or the number of binary 1's in each respective bit map.
FIG. 9 illustrates one implementation of the variable rate rebuild method, including generating the control information for the FIG. 7 data structure. The description is for one of the parity arrays; modifying the machine operations to accommodate a plurality of parity arrays can be achieved in any one of several approaches. For example, only one of the three arrays 13-15 may have a non-zero bit map; then only array indicated by the non-zero bit map is processed. If a plurality of bit maps are non-zero, priority of rebuilding in the three arrays can be based upon the least busy array, a designation of relative importance of the arrays to continued successful system operations, and the like. In any event, entry of the FIG. 9 illustrated machine operations is over path 110 from either FIG. 4 or 5, the present description assumed entry from FIG. 5. At machine step 111 whether or not a measurement period has timed out is sensed. If a measurement period did not time out, then at machine step 112 the access tally is updated (other rebuild rate indicating criteria may be updated as well). Following the update, at machine decision step 113 field 83 is examined along with an elapsed time indication. Elapsed time is computed from the time of day the last rebuild for the parity error was completed, such as indicated by numeral 89 of FIG. 7. By way of design, such last rebuild is the last rebuild by any of the methods being used, the last rebuild of either the idle or variable rate rebuild method or the last rebuild achieved by the variable rate rebuild method. If the time for rebuild has not been reached, then the FIG. 5 illustrated scanning procedure is reentered. If a rebuild is to be scheduled, then at machine step 118 the cylinder and track(s) to be rebuilt are selected. In making the cylinder and track selection, it is desired to minimize seek times to reach a track(s) to be rebuilt. For a striped mode array, since the track access mechanisms (not separately shown) of the drives in each parity array always have a common radial position over disks 70 of the respective devices 20-24 (scanning a track in a current logical cylinder comprising all physical cylinders in the parity group devices at a same radial position), a track in the current cylinder or in a cylinder having a closest radial proximity to the current cylinder is selected. Such track(s) is identified by analysis of the array's FIG. 8 bit map. This analysis is straight forward and is not further described.
For a parity array operating in the independent mode, the same general approach is used. Each of the devices in the independent mode may be scanning tracks/cylinders having different radial positions. The track/cylinder to be rebuilt is the cylinder that is located at a mean radial position between a first device in the array having its access mechanism at a radial inwardmost position and a second device having its access mechanism at a radially outwardmost position or all the devices in the array. For example, if the first device has its access mechanism for scanning track 72 of FIG. 6, the second device has its access mechanism for scanning track 73 and the other two operational devices in the parity array have their respective access mechanisms positioned radially for scanning tracks radially intermediate between tracks 72 and 73, then the cylinder radially midway between tracks 72 and 73 is examined for rebuild. If the midway cylinder has no track to be rebuilt, then adjacent cylinders are successively examined at successively increasing radial distances from the midway cylinder This determination follows examining the cylinder which is next radially outward of the midway cylinder, thence if that cylinder has no track requiring rebuild, then the next radially inward cylinder is examined, etc.
After selecting the cylinder in machine step 118, a track(s) in the selected cylinder is rebuilt in machine step 119. This rebuilding is computing the data from the corresponding physical tracks in the other devices of the array; then storing the computed data into the selected track(s). Upon completion of the rebuilding, the respective bit position(s) of the array bit map of FIG. 8 is reset to 0. Then at machine step 120 whether or not the parity array is idle and there are still tracks to be rebuilt is checked. Whenever the array is idle and a track of the parity array still needs to be rebuilt, machine steps 118 and 119 are repeated until the array is no longer idle or all rebuilds have been completed, then other machine operations are performed.
Calculation of the variable rebuild rate occurs whenever the machine step 111 indicates the measurement period has expired. In one embodiment of computing the desired rebuild rate, at machine step 125 the number of accesses to the parity array are averaged to obtain an average access rate. Such averaging allows varying the measurement period without affecting other variables. The average rate is stored in field 84. Then the access tally is reset in machine step 126, such access tally may be stored in a register 80-82 as represented by numeral 89. In machine step 127 the rebuild rate is determined by comparing the access tally values in fields 85-88 with the field 84 value. Then the rebuild rate corresponding to the threshold field 85-88 having a value least less than the field 84 value is stored in field 83 as the new rebuild rate. Remember that T-1's rebuild rate is zero. If a queue length criterion is used, then the queue length(s) are examined and the rebuild rate corresponding to the respective queue lengths is selected. Of course, algorithmic calculations can be used rather than table lookup; such calculation results are rounded to a next lowest time slice or unit value.
FIG. 10 illustrates one embodiment of the idle rebuild method. Entry into the method is from FIG. 4 or over path 129 from FIG. 5. Whether or not the parity array is idle and a track in the parity array needs a rebuild is checked at machine step 130. If the parity array is not currently idle or there is no need for a track rebuild, then the operation returns over path 135 to the caller, such as the FIG. 5 illustrated selection method. When the parity array is idle with a rebuild need, then at machine step 131 a cylinder and one of its tracks are selected for rebuild. This selection uses the aforedescribed selection method. Following selection and seeking the transducer (not shown) to the selected track, machine step 132 rebuilds the track contents. Machine step 133 then checks to see if the parity array is still idle, if yes steps 131 and 132 are repeated until either no more rebuilds are needed or the parity array becomes busy. At that point other machine operations ensue.
FIG. 11 illustrates the opportunistic rebuild method. A track access operation is initiated over path 139. For purposes of illustration, the ensuing machine step 140 is an attempted read from the track to be accessed. Any access operation may be used, such as write, format write, verify, erase and the like. Assuming a read operation, machine step 141 determines whether or not a hard (uncorrectable) error has occurred. Included in the machine step 141 operation is detection that the device containing the accessed track is already known to have failed and a rebuild is pending or already in progress for that device. If the read produced no hard error, i.e. no errors detected or a corrected error occurred, machine step 142 checks the quality of the readback operation. Since a corrected error may not be repeated, machine step 142 may not invoke the opportunistic rebuild method, choosing to proceed as OK over path 143 to other operations. If the quality requirements are not met, such as determinable by evaluating read signal quality, the type and extent of the corrected error, systems requirements (desired quality of the redundancy in the array) and the like, the opportunistic rebuild is initiated (the NO exit from step 142). Machine step 144 effects a rebuild of the currently accessed track from either machine step 141 detecting a hard error or from machine step 142. Rebuilding follows the above-described method of rebuild. Upon completing rebuilding the accessed track data in machine step 144, in machine step 145 whether or not the parity array is idle is checked. If the parity array is not idle, then other operations are performed; if the parity array is idle, then machine step 146 rebuilds a next track. Such rebuild includes selecting a cylinder and track followed by the actual rebuild method. Machine steps 145 and 146 repeat until either no more rebuilds are needed (bit maps are all zeros) or the parity array becomes active.
While the invention has been particularly shown and described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US4092732 *||May 31, 1977||May 30, 1978||International Business Machines Corporation||System for recovering data stored in failed memory unit|
|US4761785 *||Jun 12, 1986||Aug 2, 1988||International Business Machines Corporation||Parity spreading to enhance storage access|
|US4775978 *||Jan 12, 1987||Oct 4, 1988||Magnetic Peripherals Inc.||Data error correction system|
|US4855907 *||Apr 22, 1988||Aug 8, 1989||International Business Machines Corporation||Method for moving VSAM base clusters while maintaining alternate indices into the cluster|
|US4870643 *||Nov 6, 1987||Sep 26, 1989||Micropolis Corporation||Parallel drive array storage system|
|US5101492 *||Nov 3, 1989||Mar 31, 1992||Compaq Computer Corporation||Data redundancy and recovery protection|
|US5179704 *||Mar 13, 1991||Jan 12, 1993||Ncr Corporation||Method and apparatus for generating disk array interrupt signals|
|US5195100 *||Mar 2, 1990||Mar 16, 1993||Micro Technology, Inc.||Non-volatile memory storage of write operation identifier in data sotrage device|
|US5208813 *||Oct 23, 1990||May 4, 1993||Array Technology Corporation||On-line reconstruction of a failed redundant array system|
|*||CA1296103A||Title not available|
|1||Patterson et al. "A Case for Redundant Arrays of Inexpensive Disks(RAID)" ACM 1988 Mar. 1988 pp. 109-116.|
|2||*||Patterson et al. A Case for Redundant Arrays of Inexpensive Disks(RAID) ACM 1988 Mar. 1988 pp. 109 116.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US6516425 *||Oct 29, 1999||Feb 4, 2003||Hewlett-Packard Co.||Raid rebuild using most vulnerable data redundancy scheme first|
|US6959399||Sep 24, 2001||Oct 25, 2005||International Business Machines Corporation||Selective automated power cycling of faulty disk in intelligent disk array enclosure for error recovery|
|US6976187 *||Nov 8, 2001||Dec 13, 2005||Broadcom Corporation||Rebuilding redundant disk arrays using distributed hot spare space|
|US7139931 *||Nov 18, 2002||Nov 21, 2006||Aristos Logic Corporation||Method of controlling the system performance and reliability impact of hard disk drive rebuild|
|US7185222||Nov 14, 2003||Feb 27, 2007||International Business Machines Corporation||Apparatus, system, and method for maintaining data in a storage array|
|US7398418||Mar 22, 2007||Jul 8, 2008||Compellent Technologies||Virtual disk drive system and method|
|US7404102||Mar 22, 2007||Jul 22, 2008||Compellent Technologies||Virtual disk drive system and method|
|US7428691||Nov 12, 2003||Sep 23, 2008||Norman Ken Ouchi||Data recovery from multiple failed data blocks and storage units|
|US7493514||Mar 22, 2007||Feb 17, 2009||Compellent Technologies||Virtual disk drive system and method|
|US7574622||Mar 22, 2007||Aug 11, 2009||Compellent Technologies||Virtual disk drive system and method|
|US7574623||Apr 29, 2005||Aug 11, 2009||Network Appliance, Inc.||Method and system for rapidly recovering data from a “sick” disk in a RAID disk group|
|US7587630||Apr 29, 2005||Sep 8, 2009||Network Appliance, Inc.||Method and system for rapidly recovering data from a “dead” disk in a RAID disk group|
|US7613945||Aug 13, 2004||Nov 3, 2009||Compellent Technologies||Virtual disk drive system and method|
|US7650352||Mar 23, 2006||Jan 19, 2010||International Business Machines Corporation||System and method for increasing availability of an index|
|US7849352||Dec 11, 2008||Dec 7, 2010||Compellent Technologies||Virtual disk drive system and method|
|US7886111||Feb 8, 2011||Compellent Technologies||System and method for raid management, reallocation, and restriping|
|US7941695||Feb 4, 2009||May 10, 2011||Compellent Technolgoies||Virtual disk drive system and method|
|US7945810||Aug 10, 2009||May 17, 2011||Compellent Technologies||Virtual disk drive system and method|
|US7962778||Nov 2, 2009||Jun 14, 2011||Compellent Technologies||Virtual disk drive system and method|
|US8020036||Oct 30, 2008||Sep 13, 2011||Compellent Technologies||Virtual disk drive system and method|
|US8060772||Jan 4, 2008||Nov 15, 2011||International Business Machines Corporation||Storage redundant array of independent drives|
|US8230193||Feb 7, 2011||Jul 24, 2012||Compellent Technologies||System and method for raid management, reallocation, and restriping|
|US8321721||May 10, 2011||Nov 27, 2012||Compellent Technologies||Virtual disk drive system and method|
|US8468292||Jul 13, 2009||Jun 18, 2013||Compellent Technologies||Solid state drive data storage system and method|
|US8473776||Dec 6, 2010||Jun 25, 2013||Compellent Technologies||Virtual disk drive system and method|
|US8555108||May 10, 2011||Oct 8, 2013||Compellent Technologies||Virtual disk drive system and method|
|US8560880||Jun 29, 2011||Oct 15, 2013||Compellent Technologies||Virtual disk drive system and method|
|US8601035||Jun 22, 2007||Dec 3, 2013||Compellent Technologies||Data storage space recovery system and method|
|US8751861 *||Feb 8, 2012||Jun 10, 2014||Lsi Corporation||System and method for improved rebuild in RAID|
|US8819334||Jun 17, 2013||Aug 26, 2014||Compellent Technologies||Solid state drive data storage system and method|
|US9021295||Oct 7, 2013||Apr 28, 2015||Compellent Technologies||Virtual disk drive system and method|
|US9047216||Oct 14, 2013||Jun 2, 2015||Compellent Technologies||Virtual disk drive system and method|
|US9146851||Mar 26, 2012||Sep 29, 2015||Compellent Technologies||Single-level cell and multi-level cell hybrid solid state drive|
|US9244625||Jul 23, 2012||Jan 26, 2016||Compellent Technologies||System and method for raid management, reallocation, and restriping|
|US9251049||Dec 3, 2013||Feb 2, 2016||Compellent Technologies||Data storage space recovery system and method|
|US20030088803 *||Nov 8, 2001||May 8, 2003||Raidcore, Inc.||Rebuilding redundant disk arrays using distributed hot spare space|
|US20030093721 *||Sep 24, 2001||May 15, 2003||International Busiess Machines Corporation||Selective automated power cycling of faulty disk in intelligent disk array enclosure for error recovery|
|US20050055603 *||Aug 13, 2004||Mar 10, 2005||Soran Philip E.||Virtual disk drive system and method|
|US20050102552 *||Nov 18, 2002||May 12, 2005||Robert Horn||Method of controlling the system performance and reliability impact of hard disk drive rebuild|
|US20050120267 *||Nov 14, 2003||Jun 2, 2005||Burton David A.||Apparatus, system, and method for maintaining data in a storage array|
|US20070180306 *||Mar 22, 2007||Aug 2, 2007||Soran Philip E||Virtual Disk Drive System and Method|
|US20070226235 *||Mar 23, 2006||Sep 27, 2007||International Business Machines Corporation||System and Method for Increasing Availability of an Index|
|US20070234109 *||Mar 22, 2007||Oct 4, 2007||Soran Philip E||Virtual Disk Drive System and Method|
|US20070234110 *||Mar 22, 2007||Oct 4, 2007||Soran Philip E||Virtual Disk Drive System and Method|
|US20070234111 *||Mar 22, 2007||Oct 4, 2007||Soran Philip E||Virtual Disk Drive System and Method|
|US20080320061 *||Jun 22, 2007||Dec 25, 2008||Compellent Technologies||Data storage space recovery system and method|
|US20090177918 *||Jan 4, 2008||Jul 9, 2009||Bulent Abali||Storage redundant array of independent drives|
|US20090265510 *||Oct 22, 2009||Dell Products L.P.||Systems and Methods for Distributing Hot Spare Disks In Storage Arrays|
|US20130205166 *||Feb 8, 2012||Aug 8, 2013||Lsi Corporation||System and method for improved rebuild in raid|
|International Classification||G06F3/06, G11B20/18, G06F11/10, G06F11/00|
|Cooperative Classification||G06F11/1092, G11B20/1833|
|European Classification||G11B20/18D, G06F11/10R4|
|Aug 14, 2001||FPAY||Fee payment|
Year of fee payment: 8
|Aug 14, 2001||SULP||Surcharge for late payment|
Year of fee payment: 7
|Sep 16, 2005||FPAY||Fee payment|
Year of fee payment: 12
|Sep 16, 2005||SULP||Surcharge for late payment|
Year of fee payment: 11