Method and apparatus for storing data with reduced redundancy using data clusters
View Patent ↗Method and apparatus for storing data in a reduced redundancy form. Binary Large Objects (BLOBs) are partitioned into subblocks according to a partitioning method, and the subblocks are stored in subblock clusters. Each BLOB is represented as a list of spans of subblocks which identifies a contiguous sequence of subblocks within a cluster. Storage redundancy can be reduced because the spans of two different BLOBs can refer to the same subblocks. An index may be used to map subblock hashes to subblock cluster numbers.
1. A method, comprising:
dividing a Binary Large Object (BLOB) into a plurality of subblocks, where the BLOB (b) is divided by partitioning b into a plurality of subblocks, where at least one position k|k+1 in b for which b[k−A+1 . . . k+B] satisfies a predetermined constraint, where A and B are natural numbers;
storing the plurality of subblocks in a plurality of clusters, where two or more subblocks are stored in a cluster as a contiguous sequence of bytes with no intervening metadata;
creating a representation of the BLOB as one or more spans, where a span refers to a finite sequence of one or more bytes in the cluster, where a span identifies a sequence of contiguous subblocks in a cluster with a length that identifies one or more of, a number of contiguous subblocks and a number of bytes, and where a span comprises one or more of, a skip value x that indicates that the extent of the span is to be reduced by x bytes, and an extension value y that indicates that the extent of the span is to be increased by y bytes;
maintaining an index that maps the hash of at least one subblock to the cluster containing the subblock, where only the T'th subblock in a BLOB is indexed, T being a positive integer, and where the index comprises one or more hash tables; and
upon determining that a fragmentation threshold has been surpassed, duplicating a contiguous run of one or more subblocks in the store of subblocks.
2. The method of claim 1 , where the one or more spans are stored as an ordered list.
3. The method of claim 1 , where the one or more spans are stored as a tree of spans.
4. The method of claim 1 , where an upper bound is placed on one or more of, the number of subblocks in cluster, and the number of bytes in a cluster.
5. The method of claim 1 , where data structures to store the BLOB are created, but the BLOB is not stored.
6. The method of claim 1 , comprising reconstructing the BLOB from the subblocks referenced by the one or more spans.
7. The method of claim 1 , where a cluster comprises a directory of subblocks and where the directory comprises at least one of: the length of subblock, hash of subblock, position of subblock in the cluster, and an identifier for subblock.
8. The method of claim 1 , where the cluster directory is stored in the cluster.
9. The method of claim 1 , where the cluster directory is stored separately from the cluster.
10. The method of claim 1 , where the cluster directory has a fixed length.
11. The method of claim 10 , where the cluster records boundaries between contiguous runs of subblocks in the cluster.
12. The method of claim 1 , comprising compressing at least one cluster.
13. The method of claim 1 , comprising compressing at least one subblock.
14. The method of claim 1 , comprising compressing two or more adjacent subblocks.
15. The method of claim 1 , comprising maintaining an index that maps at least one subblock to the cluster containing the subblock.
16. The method of claim 1 , where the index stores the position of subblock in the cluster containing the subblock.
17. The method of claim 1 , the index comprising a digital search tree whose keys are subblock hashes.
18. The method of claim 1 , the index comprising a Btree.
19. The method of claim 1 the index comprising one or more hash tables.
20. The method of claim 1 , where a hash table entry for a subblock comprises portion of the hash of the subblock.
21. The method of claim 1 , the hash table comprising buckets.
22. The method of claim 1 , comprising checking for duplicate subblocks by checking the index before adding a subblock to a cluster.
23. The method of claim 1 , comprising checking for duplicate subblocks by comparing the hashes of subblocks to be stored with the hashes of at least one of the subblocks in a cluster where an index indicates a subblock is stored.
24. The method of claim 1 , where a span identifies a subblock using a portion of the hash of the subblock.
25. The method of claim 1 , comprising duplicating a contiguous run of less than T present subblocks in the store of subblocks, where T is a predefined threshold of subblocks.
26. The method of claim 25 , T being two.
27. The method of claim 1 , comprising duplicating a contiguous run of one or more subblocks in the store of subblocks.
28. The method of claim 1 , comprising augmenting at least one span X with an alternative span that refers to a copy of the data referred to by span X.
29. The method of claim 1 , comprising: upon determining that the location of a subblock X as a function of the index, searching forwards from subblock X to find the longest matching run of subblocks with the subblocks being stored.