Out-of core similarity matching

Patent metadata

Field	Value
Publication number	US-9727573-B1
Application number	US-201414482589-A
Country	US
Kind code	B1
Filing date	Sep 10, 2014
Priority date	Dec 22, 2011
Publication date	Aug 8, 2017
Grant date	Aug 8, 2017

Abstract

A method for storing data in a data storage system by partitioning the data into a plurality of data chunks and generating representative data for each of the plurality of chunks by applying a predetermined algorithm to each chunk of the plurality of chunks. Subsequently, the representative data is compared and sorted. Representative data for base data chunks and representative data for other data chunks that can be stored relative to the base data chunks are identified by evaluating the sorted set of representative data. Finally, each of the other data chunks identified as those that can be stored relative to a base data chunk are stored in the data storage system as the difference between the data chunk and a base data chunk.

First claim

What is claimed is:

1. A computer-implemented method for data reduction, the method comprising:
in response to a request for compressing data in a data storage system, partitioning the data into a plurality of data chunks, including a target data chunk and a base data chunk;
generating representative data for the target data chunk and the base data chunk by applying a predetermined algorithm to the target data chunk and the base data chunk, the representative data including fingerprints of the target data chunk and the base data chunk and a plurality of features extracted from the target data chunk and the base data chunk, wherein each of the plurality of features is a value having a property that a probability of the target data chunk having same representative value as the base data chunk is proportional to data similarity of the target data chunk and the base data chunk;
sorting the representative data for the target data chunk and the base data chunk to form a sorted representative data list based on a first feature defined in the representative data for the target data chunk and the base data chunk; and
generating a delta data chunk as the difference between the target data chunk and the base data chunk where the representative data of the target chunk is proximate to the representative data of the base data chunk in the sorted representative data list.

2. The computer-implemented method of claim 1 , wherein the representative data of the base data chunk is proximate to the representative data of the target data chunk such that the representative data of the base data chunk has one or more features that are identical to one or more corresponding features of the representative data of the target data chunk.

3. The computer-implemented method of claim 1 , wherein the base data chunk is chosen based on an age comparison with the target data chunk.

4. The computer-implemented method of claim 1 , wherein the base data chunk is chosen based on physical locality in the data storage system in relation to the target data chunk.

5. The computer-implemented method of claim 1 , further comprising: storing the delta data chunk and the base data chunk in the data storage system, wherein the delta data chunk and the base data chunk represent the target data chunk.

6. The computer-implemented method of claim 5 , further comprising: removing the target data chunk from the data storage system after the delta data chunk is generated.

7. The computer-implemented method of claim 1 , further comprising: transmitting the delta data chunk and the base data chunk to an auxiliary data storage system.

8. The computer-implemented method of claim 1 , further comprising: estimating the compression achievable of delta encoding the target data chunk relative to the base data chunk.

9. The computer-implemented method of claim 1 , wherein sorting the representative data comprises:
dividing representative data of the plurality of data chunks into a plurality of bin files where representative data of the plurality of data chunks sharing common representative data are placed in a same bin file, each of the plurality of bin files sized to fit within a main memory of the data storage system;
reading into the main memory each of the plurality of bin files; and
comparing and sorting each of the plurality of bin files according to a first feature defined in representative data of the plurality of data chunks.

10. The computer-implemented method of claim 9 , wherein sorting the representative data occurs for each feature of the representative data such that the representative data in the representative data list is first sorted based on the first feature of the plurality of data chunks and subsequently sorted based on a second feature of the plurality of data chunks, and
wherein during the sorting iteration using the first feature delta data chunks are generated for one or more pairs of data chunks that have identical first features and wherein during the sorting iteration using the second feature delta data chunks are generated for one or more pairs of data chunks that have identical second features.

11. The computer-implemented method of claim 10 , further comprising:
removing, following generation of delta data chunks for one or more pairs of data chunks that have equal first features and prior to sorting the representative data list based on the second feature, the representative data from the representative data list for one of the data chunks in each pair of data chunks.

12. The computer-implemented method of claim 9 , wherein representative data of the plurality of data chunks in the same bin file share the first feature,
wherein sorting the representative data occurs in each bin file using the first feature of the plurality of data chunks and secondarily based on a second feature of the plurality of data chunks, and
wherein delta data chunks are generated for one or more pairs of data chunks that have one or more of identical first features or identical second features.

13. The computer-implemented method of claim 12 , further comprising:
sorting, following generating the delta data chunks for one or more pairs of data chunks that have identical first or second features, the representative data using the second feature; and
generating one or more pairs of delta data chunks for data chunks that have representative data with identical second features.

14. The computer-implemented method of claim 1 , wherein generating representative data comprises:
inputting a data chunk into a collision-resistant hash function;
receiving from the hash function a hash value for the data chunk; and
assigning the hash value to the data chunk.

15. The computer-implemented method of claim 14 , wherein generating representative data further comprises:
extracting one or more features from each of the plurality of data chunks; and
assigning the features to each of the plurality of data chunks so that representative data includes the one or more features and the hash value.

16. The computer-implemented method of claim 1 , wherein the predetermined algorithm extracts one or more features from each of the plurality of data chunks.

17. A non-transitory computer-readable storage medium having instructions stored therein, which when executed by a computer, cause the computer to:
in response to a request for compressing data in a data storage system, partition the data into a plurality of data chunks, including a target data chunk and base data chunk;
generate representative data for the target data chunk and the base data chunk by applying a predetermined algorithm to the target data chunk and the base data chunk, the representative data including fingerprints of the target data chunk and the base data chunk and a plurality of features extracted from the target data chunk and the base data chunk, wherein each of the plurality of features is a value having a property that a probability of the target data chunk having same representative value as the base data chunk is proportional to data similarity of the target data chunk and the base data chunk;
sort the representative data for the target data chunk and the base data chunk to form a sorted representative data list based on a first feature defined in the representative data for the target data chunk and the base data chunk; and
generate a delta data chunk in the data storage system as the difference between the target data chunk and the base data chunk where the representative data of the target chunk is proximate to the representative data of the base data chunk in

Assignees

• Emc Ip Holding Co Llc

Inventors

• Wallace Grant
• Shilane Philip N
• Douglis Frederick

Classifications

• G06F16/22

  Indexing; Data structures therefor; Storage structures · CPC title

• G06F16/174

  Redundancy elimination performed by the file system (error detection or correction of the data by redundancy in operations G06F11/14) · CPC title

• G06F3/0608

  Saving storage space on storage systems · CPC title

• G06F16/1752

  based on file chunks · CPC title

• G06F16/24556

  Aggregation; Duplicate elimination · CPC title