Map-reduce with merge to process multiple relational datasets

• US 7,523,123 B2
• Filed: 11/16/2006
• Issued: 04/21/2009
• Est. Priority Date: 11/16/2006
• Status: Active Grant

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Yang, Hung-Chih, Dasdan, Ali, Hsiao, Ruey-Lung
• Primary Examiner(s): Le; Uyen T.
• Application Number:
  US11/560,523
• Publication Number:
  US 20080120314A1
• Time in Patent Office: 887 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/2456 Join operations
• CPC Class Codes: Y10S 707/99942 Manipulating data structure, e.g. compression, compaction, compilation
• CPC Class Codes: Y10S 707/99943 Generating database or data structure, e.g. via user interface
• CPC Class Codes: Y10S 707/99945 Object-oriented database structure processing

Abstract

A method of processing relationships of at least two datasets is provided. For each of the datasets, a map-reduce subsystem is provided such that the data of that dataset is mapped to corresponding intermediate data for that dataset. The intermediate data for that dataset is reduced to a set of reduced intermediate data for that dataset. Data corresponding to the sets of reduced intermediate data are merged, in accordance with a merge condition. In some examples, data being merged may include the output of one or more other mergers. That is, generally, merge functions may be flexibly placed among various map-reduce subsystems and, as such, the basic map-reduce architecture may be advantageously modified to process multiple relational datasets using, for example, clusters of computing devices.

Claims

1. A method of processing data and data relationships of at least two datasets, comprising:

for the data of each one of the datasets,

mapping the data of that dataset to corresponding intermediate data for that dataset; and

reducing the intermediate data for that dataset to a set of reduced intermediate data for that dataset; and

merging data corresponding to the sets of reduced intermediate data, in accordance with a merge condition.

2. The method of claim 1, wherein:

each set of mapped intermediate data includes a set of tuples;

each set of reduced intermediate data includes a set of tuples;

the merging step includes executing a selector task that determines, based at least in part on the merge condition, which tuples of the sets of reduced intermediate data to merge, the merged output includes a set of tuples that are produced based on the reduced intermediate data.

3. The method of claim 1, wherein:

each set of mapped intermediate data includes a set of tuples, each tuple characterized as a collection of attributes;

each set of reduced intermediate data includes a set of tuples, each tuple characterized as a collection of attributes;

the merge condition includes a relationship between keys characterizing the tuples.

4. The method of claim 3, further comprising:

configuring the merge condition.

5. The method of claim 3, wherein:

the relationship between keys of the tuples includes a key of a tuple of one set of reduced intermediate data being equivalent to a key of a tuple of another set of reduced intermediate data.

6. The method of claim 1, wherein:

each set of mapped intermediate data includes a set of tuples, each tuple characterized by a key/value pair, wherein the keys and values are characterized as sets of attributes;

each set of reduced intermediate data includes a set of tuples, each tuple characterized by a key/value pair, wherein the keys and values are characterized as sets of attributes.

7. The method of claim 6, wherein:

the merging step includes merging the tuples for which the keys are equivalent based on conditions defined by a user-defined procedure.

8. The method of claim 1, wherein:

the at least two datasets include a first dataset and a second dataset but do not include a third dataset;

the merging step is a first merging step; and

the method further comprises;

mapping the data of the third dataset to corresponding intermediate data for the third dataset; and

reducing the intermediate data for the third dataset to a set of reduced intermediate data for the third dataset; and

a second merging step, comprising merging data corresponding to the set of reduced intermediate data for the third dataset with data corresponding to an output of the first merging step.

9. The method of claim 1, wherein:

the at least two datasets include a first dataset and a second dataset but do not include a third dataset and a fourth dataset;

the merging step is a first merging step; and

the method further comprises;

for the data of each of third and fourth datasets,

mapping the data of that dataset to corresponding intermediate data for that dataset; and

reducing the intermediate data for that dataset to a set of reduced intermediate data for that dataset; and

a second merging step, comprising merging the set of reduced intermediate data for the third dataset with the set of reduced intermediate data for the fourth dataset; and

a third merging step, comprising merging data corresponding to an output of the first merging step with data corresponding to an output of the second merging step.

10. The method of claim 9, further comprising:

mapping the output of the first merging step to intermediate data for the output of the first merging step; and

reducing the intermediate data for the output of the first merging step to a set of reduced intermediate data for the output of the first merging step;

wherein the reduced intermediate data for the output of the first merging step is the data corresponding to an output of the first merging step.

11. The method of claim 1, wherein:

the mapping, reducing and merging steps include processing metadata of the data processed at each respective one of the steps.

12. The method of claim 11, wherein:

the metadata processed at each of the mapping, reducing and merging steps indicates characteristics, types and attributes of the data processed at that step.

13. The method of claim 11, wherein:

the metadata processed at each step is provided to another step in a pipelined fashion.

14. The method of claim 1, wherein:

data processed at the mapping step is read from at least one of the group consisting of;

a plurality of file system sources;

a plurality of data repository sources;

mapping step intermediate outputs;

reducing step intermediate outputs; and

merging step intermediate outputs.

15. The method of claim 1, wherein:

at least some of the mapping, reducing and merging steps occur on a single computing device such that data being shared between such steps need not be stored to disk.

16. The method of claim 15, wherein:

the merging step is combined with a succeeding mapping step, thereby saving a disk access by the succeeding mapping step.

17. The method of claim 15, wherein:

the reducing step is combined with a succeeding mapping step, thereby saving a disk access by the succeeding mapping step.

18. The method of claim 15, wherein:

the reducing step for one dataset is combined with the merging step, and

the merging step reads intermediate remotely from a reducer for another dataset.

19. A computing system configured to process relationships of at least two datasets, the computing system including at least one computing device configured to:

for the data of each one of the datasets,

map the data of that dataset to corresponding intermediate data for that dataset; and

reduce the intermediate data for that dataset to a set of reduced intermediate data for that dataset; and

merge data corresponding to the sets of reduced intermediate data, in accordance with a merge condition.

20. The computing system of claim 19, wherein:

the at least one computing device being configured to map the data of the data set to corresponding intermediate data for that dataset includes each of a first portion of a plurality of the at least one computing devices being configured to execute a separate mapping process, in parallel;

the at least one computing device being configured to reduce the data of the data set to a set of reduced intermediate data for that dataset includes each of a second portion of a plurality of the at least one computing devices being configured to execute a separate reducing process, in parallel;

the at least one computing device being configured to merge the sets of the reduced intermediate data includes each of a third portion of a plurality of the at least one computing devices being configured to execute a separate merge process, in parallel.

21. The computing system of claim 20, wherein:

each of the first portion of the plurality of computing devices is configured to produce intermediate data that is partitioned into intermediate data input files to be provided to every one of the second portion of computing devices;

each of the second portion of the plurality of computing devices is configured to receive all of the intermediate data input files; and

each of the third portion of the plurality of computing devices is configured to receive selected outputs from at least two of the second portion of the plurality of computing devices.

22. The computing system of claim 21, wherein:

each of at least some of the second portion of computing devices is configured to execute the separate reducing process in response to at least a partial set of the intermediate data input files being provided; and

each of at least some of the third portion of computing devices is configured to execute the separate merging process in response to at least a partial set of the selected outputs from at least two of the second portion of the plurality of computing devices being ready.

23. The computing system of claim 19, wherein:

the map, reduce and merge are computed in a pipelined fashion.

24. The computing system of claim 19, wherein:

the at least one computing device being configured to map the data of the data set to corresponding intermediate data for that dataset includes the at least one computing devices being configured to execute a plurality of mapping processes;

the at least one computing device being configured to reduce the data of the data set to a set of reduced intermediate data for that dataset includes the at least one computing device being configured to execute a plurality of reducing processes;

the at least one computing device being configured to merge the sets of the reduced intermediate data includes the at least one computing device being configured to execute a plurality of merge processes;

each of the reducing processes is configured to reduce data, of the intermediate data input files, for a partition to which that reducing process corresponds; and

each of the merging processes is configured to merge data, of the selected outputs from at least two of the reducing processes, for a partition to which that merging process corresponds.

25. The computing system of claim 24, wherein:

each of the mapping processes is configured to partition the produced intermediate data based on a hash process applied to the intermediate data and to store each partition of the intermediate data in a file that corresponds to a result of the applied hash process; and

each of the merging processes being configured to merge data for a partition to which that merging process corresponds includes merging data for partitions based on the result of the applied hash process.

26. The computing system of claim 24, wherein:

each of the mapping processes is configured to partition the produced intermediate data based on a determination of which range of values the intermediate data belongs and to store each partition of the intermediate data in a file that corresponds to the range to which the value of the intermediate data is determined to belong; and

each of the merging processes being configured to merge data for a partition to which that merging process corresponds includes merging data for partitions based on the result of the range determining process.

27. The computing system of claim 24, wherein each of the merging processes is configured to:

read from at least two sets of reduced intermediate data; and

select which subsets from the at least two sets of reduced intermediate data to merge, using a reducing process identification that uniquely identifies each subset of reduced intermediate data with a reducing process from which that subset of reduced intermediate data is provided.

28. The computing system of claim 19, wherein:

the at least one computing device is configured such that the mapping, reducing and merging collectively accomplish a relational operation.

29. The computing system of claim 28, wherein the at least one computing device is configured such that the relational operation is a relational algebra operation.

30. The computing system of claim 29, wherein the at least one computing device is configured with the merge condition so as to merge only records with same keys from the sets of reduced intermediate data, to accomplish a relational algebra “

natural join”

or “

equijoin”

operation.

31. The computing system of claim 29, wherein the at least one computing device is configured with the merge condition corresponding to a particular algebraic condition, such that only data related by the particular algebraic condition are included in the merged data, to accomplish a “

theta join”

operation.

32. The computing system of claim 29, wherein the at least one computing device is configured with the merge condition such that all of the reduced intermediate data is included in the merged data, to accomplish a “

union”

operation.

33. The computing system of claim 29, wherein the at least one computing device is configured with the merge condition corresponding to a particular algebraic condition, such that only data that is present in both of two subsets of the reduced intermediate data is included in the merged data, to accomplish an “

intersection”

operation.

34. The computing system of claim 29, wherein the at least one computing device is configured with the merge condition corresponding to a particular algebraic condition, such that only data that is present in one subset of the reduced intermediate data, that is not included in a second subset of the reduced intermediate data, is included in the merged data, to accomplish a “

difference”

relational algebra operation or a database query language “

not in” and

“

not exists”

operation.

35. The computing system of claim 29, wherein the at least one computing device is configured with the merge condition corresponding to a particular algebraic condition, such that one tuple of the merged output data is one tuple present in one subset of the reduced intermediate data combined with every tuple present in a second subset of the reduced intermediate data, to accomplish a “

Cartesian product”

relational algebra operation.

36. The computing system of claim 28, wherein the at least one computing device is configured such that the relational operation is a relational join operation.

37. The computing system of claim 36, wherein the at least one computing device is configured, to accomplish a hash join operation, to:

map the data of the input data sets to accomplish hash partitioning of the input data sets;

reduce to at least sort and group partitioned intermediate data with the same key values; and

merge same-key data from at least two reduced sets of intermediate data.

38. The computing system of claim 36, wherein the at least one computing device is configured, to accomplish a sort-merge join operation, to:

map the data of the input data sets to accomplish range partitioning of the input data sets;

reduce to at least sort and group partitioned intermediate data with the same key values; and

merge same-key data from at least two reduced sets of intermediate data.

39. The computing system of claim 36, wherein the at least one computing device is configured, to accomplish a nested-loop join operation, to:

map the data of the input data sets to accomplish partitioning of the input data sets;

reduce to at least sort and group partitioned intermediate data with the same key values; and

merge same-key data from at least two reduced sets of intermediate data.

40. A method of configuring a computing system to process data and data relationships of at least two datasets, comprising:

for the data of each one of the datasets,

configuring the computing system to include a mapping function to map data of that dataset to corresponding intermediate data for that dataset; and

configuring the computing system to include a reducing function to reduce the intermediate data for that dataset to a set of reduced intermediate data for that dataset; and

configuring the computing system to include a merging function to merge data corresponding to the sets of reduced intermediate data, in accordance with a merge condition.

41. The method of claim 40, further comprising:

configuring the computing system to organize the mapping, reducing and merging function to operate in a pipelined fashion.

42. The method of claim 41, wherein:

configuring the computing system to organize the mapping, reducing and merging function to operate in a pipelined fashion includes configuring a direction of dataflow among the mapping, reducing and merging functions.

43. The method of claim 41, wherein:

configuring the computing system to organize the mapping, reducing and merging function to operate in a pipelined fashion includes configuring a source and destination of data processed by the mapping, reducing and merging functions.

44. The method of claim 41, wherein:

the configuring is according to a configuration provided by a user.

45. The method of claim 41, wherein:

the configuring is according to a configuration at least part of which is determined automatically.

46. A computer program product for processing data and data relationships of at least two datasets, the computer program product comprising at least one computer-readable medium having computer program instructions stored therein which are operable to cause at least one computing device to:

for the data of each one of the datasets,

map the data of that dataset to corresponding intermediate data for that dataset; and

reduce the intermediate data for that dataset to a set of reduced intermediate data for that dataset; and

merge data corresponding to the sets of reduced intermediate data, in accordance with a merge condition.

47. The computer program product of claim 46, wherein:

each set of mapped intermediate data includes a set of tuples;

each set of reduced intermediate data includes a set of tuples;

the merging computer program instructions includes computer program instructions operable to cause the at least one computer device to execute a selector task that determines, based at least in part on the merge condition, which tuples of the sets of reduced intermediate data to merge, the merged output includes a set of tuples that are produced based on the reduced intermediate data.

48. The computer program product of claim 46, wherein:

each set of mapped intermediate data includes a set of tuples, each tuple characterized as a collection of attributes;

each set of reduced intermediate data includes a set of tuples, each tuple characterized as a collection of attributes;

the merge condition includes a relationship between keys characterizing the tuples.

49. The computer program product of claim 48, further comprising:

computer program instructions operable to cause the at least one computer device to configure the merge condition.

50. The computer program product of claim 48, wherein:

the relationship between keys of the tuples includes a key of a tuple of one set of reduced intermediate data being equivalent to a key of a tuple of another set of reduced intermediate data.

51. The computer program product of claim 46, wherein:

each set of mapped intermediate data includes a set of tuples, each tuple characterized by a key/value pair, wherein the keys and values are characterized as sets of attributes;

each set of reduced intermediate data includes a set of tuples, each tuple characterized by a key/value pair, wherein the keys and values are characterized as sets of attributes.

52. The computer program product of claim 51, wherein:

the merging computer program instructions includes computer program instructions operable to cause the at least one computer device to merge the tuples for which the keys are equivalent based on conditions defined by a user-defined procedure.

53. The computer program product of claim 46, wherein:

the at least two datasets include a first dataset and a second dataset but do not include a third dataset;

the merging computer program instructions are first merging instructions; and

the computer program instructions are further operable to cause at least one computing device to;

map the data of the third dataset to corresponding intermediate data for the third dataset; and

reduce the intermediate data for the third dataset to a set of reduced intermediate data for the third dataset; and

wherein the computer program instructions further include second merging instructions operable to cause the at least one computing device to merge data corresponding to the set of reduced intermediate data for the third dataset with data corresponding to an output of the first merging instructions.

54. The computer program product of claim 46, wherein:

the at least two datasets include a first dataset and a second dataset but do not include a third dataset and a fourth dataset;

the merging computer program instructions are first merging instructions; and

the computer program instructions are further operable to cause at least one computing device to;

for the data of each of third and fourth datasets,

map the data of that dataset to corresponding intermediate data for that dataset; and

reduce the intermediate data for that dataset to a set of reduced intermediate data for that dataset;

wherein the computer program instructions further include second merging instructions operable to cause the at least one computing device to merge the set of reduced intermediate data for the third dataset with the set of reduced intermediate data for the fourth dataset; and

wherein the computer program instructions further include second merging instructions operable to cause the at least one computing device to merge data corresponding to an output of the at least one computing device operating according the first merging instructions with data corresponding an output of the at least one computing device operating according to an output of the second merging instructions.

55. The computer program product of claim 54, wherein the computer program instructions are further operable to cause at least one computing device to:

map the output of the first merging step to intermediate data for the output of the at least one computing device operating according the first merging instructions; and

reduce the intermediate data for the output of the at least one computing device operating according the first merging instructions to a set of reduced intermediate data for the output of the at least one computing device operating according the first merging instructions;

wherein the reduced intermediate data for the output of the at least one computing device operating according to the first merging instructions is the data corresponding to an output of the at least one computing device operating according to the first merging instructions.