System and method for identifying inconsistent parity in an array of storage

• US 5,574,882 A
• Filed: 03/03/1995
• Issued: 11/12/1996
• Est. Priority Date: 03/03/1995
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Riegel, Geoffrey A., Wyllie, James C., Menon, Jaishankar M.
• Primary Examiner(s): Swann, Tod R.
• Assistant Examiner(s): Thai, Tuan V.
• Application Number:
  US08/397,817
• Time in Patent Office: 620 Days
• Field of Search: 395/438, 395/439, 395/441, 395/471, 395/472, 395/479, 395/481, 395/490, 395/491, 395/182.04, 395/183.18, 395/404
• US Class Current: 711/114
• CPC Class Codes: G06F 11/1076 Parity data used in redundant arrays of independent storages, e.g. in RAID systems
• CPC Class Codes: G06F 2211/1009 Cache, i.e. caches used in RAID system with parity

Abstract

A system and method are provided that is used by software implemented Redundancy Array of Inexpensive Disk (RAID) arrays to achieve adequate performance and reliability, as well as to improve performance or low cost hardware Raids. The enhancements to the basic RAID implementation speeds up recovery time for software RAIDS. A method is provided for storing data in an array of storage devices. A plurality of block locations on the storage devices are logically arranged as a parity group wherein a parity block stored in a block location as part of a parity group is logically derived from the combination of data blocks stored in the parity group, and each block in a parity group is stored on a different storage device. A plurality of parity groups are grouped into a parity group set. A request is received to write a new data block location on a storage device. The old data block stored at the block location is read. The new data block is written to the block location. When the parity set is in an unmodified state prior to the current write, an indicator is written to the storage device that the parity group set is in a modified state. In a preferred embodiment, this enhancement uses a bit map stored on disk, called Parity Group Set, (PGS) bit map, to mark inconsistent parity groups, replacing the Non-Volatile Random Access Memory, (NVRAM) used for similar purposes by hardware RAIDs. Further enhancements optimized sequential input/output, (I/O) data stream.

Claims

1. A method for storing data in an array of storage devices, including processor and memory comprising the steps of:

(A) logically arranging a plurality of block locations on the storage devices as a plurality of parity groups wherein a parity block stored in a block location as part of a parity group is logically derived from the combination of data blocks stored in the block locations of the parity group, and each parity and data block in a parity group is stored on a different storage device;

(B) grouping a set of the plurality of parity groups into a parity group set; and

(C) when writing a new data block to a data block location on a storage device;

(i) reading an old data block stored at the data block location;

(ii) writing the new data block to the data block location;

(iii) identifying a parity group and a parity group set to which the block location belongs;

(iv) only when the identified parity group set is in an unmodified state prior to writing the new data block, writing to the storage device an indicator that the identified parity group set is in a modified state indicating that during a recovery process, all parity groups in the identified parity group set will be checked for inconsistent parity; and

(D) using a parity group set counter to track whether the parity group set is in an unmodified state and when there are no longer any write operations currently modifying the parity group set, an indicator is set that the parity group set is in an unmodified state.

2. The method for storing data in an array of storage devices as called for in claim 1, further comprising the steps of:

(D) reading an old parity block for the identified parity group to which the data block location belongs;

(E) calculating a new parity block for the identified parity group from the old data block, the old parity block and the new data block;

(F) writing the new parity block to a location on the storage device for the old parity block;

(G) after completing step (F), indicating that the identified parity group is in an unmodified state;

(H) determining whether all parity groups in the identified parity group set are in an unmodified state;

(I) for each parity group set where all parity groups in the parity group set are in unmodified states, writing in memory an indicator that the parity group set is in an unmodified state; and

(J) occasionally writing to a storage device, indicators of which parity group sets are in an unmodified state, wherein, during a recovery process, no parity group in a parity group set having an indicator written to the storage device indicating the parity group set is in an unmodified state will be checked for inconsistent parity.

3. The method for storing data in an array of storage devices as called for in claim 2, further comprising the steps of:

locking the data block location before step (C)(i);

unlocking the data block location after step (C) (ii);

locking a block location of the old parity block before step (D); and

unlocking the block location of the old parity block after step (F).

4. The method for storing data in an array of storage devices as called for in claim 1, further comprising the steps of:

(D) calculating a change parity block for the identified parity group from the old data block and the new data block;

(E) reading an old parity block for the identified parity group to which the data block location belongs;

(F) calculating a new parity block from the change parity block and the old parity block; and

(G) writing the new parity block to a location on the storage device for the old parity block;

(H) after completing step (G), indicating that the identified parity group is in an unmodified state;

(I) determining whether all parity groups in the identified parity group set are in an unmodified state;

(J) for each parity group set where all parity groups in the parity group set are in unmodified states, writing in memory an indicator that the parity group set is in an unmodified state; and

(K) occasionally writing to a storage device, indicators of which parity group sets are in an unmodified state, wherein, during a recovery process, no parity group in a parity group set having an indicator written to the storage device indicating that the parity group set is in an unmodified state will be checked for inconsistent parity.

5. The method for storing data in an array of storage devices as called for in claim 1, wherein each parity group set is represented in a bit map having a bit set when the parity group set is in a modified state and said bit map is stored in memory and on a storage device and wherein the bit is reset when there are no longer any write operations modifying the parity group.

6. A method for recovering from a failure in a storage system using the method for storing data in an array of storage devices as called for in claim 5, comprising the steps of:

(A) retrieving from a storage device the parity group set bit map;

(B) for each parity group set having a bit set, identifying each inconsistent parity group in each said parity group set having a bit set, having an inconsistent parity block; and

(C) for each said inconsistent parity group, updating the parity block for the inconsistent parity group based on the data blocks of the inconsistent parity group.

7. A method for recovering from a failure as called for in claim 6 further comprising the steps of:

writing the parity group set bit map in two locations on the storage devices;

writing a time stamp at the beginning and end of the bit map as the bit map is written to the storage devices;

retrieving from the storage devices the most recent consistent time stamped version of the bit map when retrieving the bit map for a recovery process.

8. A method for recovering from a failure in a storage system using the method for storing data in an array of storage devices as called for in claim 1, comprising the steps of:

(A) reading from a storage device, indicators identifying modified parity group sets;

(B) for each modified parity group set, identifying each inconsistent parity group in said modified parity group set having an inconsistent parity block; and

(C) for each said inconsistent parity group, updating the parity block for the inconsistent parity group based on the data blocks of the inconsistent parity group.

9. The method for storing data in an array of storage devices as called for in claim 1 further comprising the steps of:

(D) determining whether the new data block is a continuation of a sequential data stream;

(E) calculating a change parity block for the identified parity group from the old data block and the new data block; and

(F) when said new data block is not the continuation of said sequential data stream;

(i) reading an old parity block for each parity group to which each block location that is part of the sequential data stream belongs;

(ii) calculating a new parity block for each parity group to which each block location that is part of the sequential data stream belongs based on a change parity block for said each parity group and the old parity block for said each parity group; and

(iii) writing the new parity block for each parity group to which each block location that is part of the sequential data stream belongs to a block location for the old parity block of said each parity group.

10. The method for storing data in an array of storage devices as called for in claim 9 wherein the indicator that the identified parity group set is in a modified state is written to the storage device after the new data block is determined to be in a parity group that is in a new parity group set from the parity group set of the previous data block of the data stream.

11. The method for storing data in an array of storage devices as called for in claim 10, further comprising the step of:

(iv) after completing step (F)(iii), for each parity group set in an unmodified state, writing in memory an indicator that the parity group set is in the unmodified state.

12. In a storage array system including plurality of storage devices, each storage device comprising a plurality of block locations, wherein a group of block locations are designated as a parity group, each parity group including a plurality of data blocks and a parity block stored in the parity group block locations, the parity block being logically derived from the combination of data blocks of the parity group, a method for writing a new data block to a designated block location in an array of storage devices without first storing the data blocks and parity block in a cache memory, comprising the ordered steps of:

(A) determining an old parity block corresponding to a parity group of the designated block location;

(B) locking the designated block location;

(C) reading an old data block stored at the designated block location;

(D) writing the new data block to the designated block location;

(E) unlocking the designated block location;

(F) locking a block location of the old parity block for the designated parity group to which the designated block location belongs;

(G) reading the old parity block;

(H) calculating a new parity block for the designated parity group based on the old data block, the old parity block and the new data block;

(I) writing the new parity block to the block location of the old parity block; and

(J) unlocking the block location of the old parity block.

13. A storage array system comprising:

a plurality of storage devices, wherein each storage device comprises a plurality of block locations;

a plurality of parity groups logically organized from said plurality of block locations on the storage devices, said parity groups comprising a parity block stored in one of the parity group block locations, said parity block logically derived from a combination of the data blocks stored in the parity group block locations, wherein each data and parity block in a parity group is stored on a different storage device;

a plurality of parity group sets, each parity group set comprising a set of said plurality of parity groups;

means for writing a new data block to one of said parity group block locations on one of said storage devices;

means for reading an old data block stored at said one of said parity group block locations;

means for identifying one of said parity group sets to which said one of said parity group block locations belongs;

means for determining when said one of said parity group sets does not have an indicator that said one of said parity group sets is in an unmodified state; and

means for writing to the storage devices, an indicator that said one of said parity group sets is in a modified state.

14. The system of claim 13 further comprising:

means for determining the parity group to which the parity group block location belongs;

means for reading an old parity block for the determined parity group;

means for calculating a new parity block for the determined parity group from the old data block, the old parity block and the new data block;

means for writing the new parity block to a parity group block location on the storage device where the old parity block was stored;

means for determining each parity group set in an unmodified state; and

means for writing to the storage device an indicator that at least one of said parity group sets is in an unmodified state.

15. The system of claim 14 wherein said indicator that at least one of said parity group sets is in an unmodified state and a modified state comprises a parity group set bit map having a bit set for each parity group set in said modified state;

and a set counter for each parity group set of the number of write operations currently modifying a parity group set wherein when said set counter becomes zero, the bit for the parity group set is changed.

16. The system of claim 13 further comprising:

means for identifying a modified parity group set;

means for identifying any parity groups in said modified parity group set having an inconsistent parity block; and

means for updating the parity block for each inconsistent parity group based on the data blocks of the parity group.

17. The system of claim 13 further comprising:

means for determining whether the new data block is a continuation of sequential data stream;

means for calculating a change parity block for a parity group from the old data block and the new data block;

means for reading an old parity block for each parity group to which a block location that is part of the sequential data stream belongs;

means for calculating a new parity block for each parity group to which a block location that is part of the sequential data stream belongs based on the change parity for each parity group and the old parity block for each parity group; and

means for writing the new parity block for each parity group to which a block location that is part of the sequential data stream belongs to each block location on the storage device of the old parity block for the parity group.

18. An article of manufacture for use in a computer system for storing data in a computer system having an array of storage devices, the computer system having means to write data blocks to the storage devices, said article of manufacture comprising a computer-readable storage medium having a computer program code embodied in said medium which may cause the computer to:

(A) logically arrange a plurality of block locations on the storage devices as a plurality of parity groups wherein a parity block stored in a block location as part of a parity group is logically derived from the combination of data blocks stored in the block locations of the parity group, and each parity and data block in a parity group is stored on a different storage device;

(B) group a set of the plurality of parity groups into a parity group set; and

(C) when writing a new data block to a data block location on a storage device;

(i) read an old data block stored at the data block location;

(ii) write the new data block to the data block location;

(iii) identify a parity group and a parity group set to which the block location belongs;

(iv) only when the identified parity group set is in an unmodified state prior to writing the new data block, write to the storage device an indicator that the identified parity group set is in a modified state so that during a recovery process, all parity groups in the identified parity group set will be checked for inconsistent parity;

(v) incrementing a parity group set counter for the identified parity group set when the identified parity group becomes modified;

(vi) decrementing the parity group set counter for the identified parity group when the identified parity group becomes unmodified; and

(vii) when the parity group set counter for the parity group set becomes zero, setting an indicator that the parity group set is in an unmodified state.

19. The new article of manufacture as claimed in claim 18 wherein the computer program code may further cause the computer to:

(D) read an old parity block for the identified parity group to which the data block location belongs;

(E) calculate a new parity block for the identified parity group from the old data block, the old parity block and the new data block;

(F) write the new parity block to a location on the storage device for the old parity block;

(G) after completing step (F), indicate that the identified parity group is in an unmodified state;

(H) determine whether all parity groups in the identified parity group set are in an unmodified state;

(I) for each parity group set where all parity groups in the parity group set are in unmodified states, write in memory an indicator that the parity group set is in an unmodified state; and

(J) intermittently write to a storage device, indicators of which parity group sets are in an unmodified state, wherein, during a recovery process, no parity group in a parity group set having an indicator written to the storage device indicating the parity group set is in an unmodified state will be checked for inconsistent parity unmodified state.