Many task computing with message passing interface

• US 10,657,107 B1
• Filed: 12/29/2019
• Issued: 05/19/2020
• Est. Priority Date: 02/05/2016
• Status: Active Grant

• Current Assignee: SAS Institute Incorporated
• Original Assignee: SAS Institute Incorporated
• Inventors: Bequet, Henry Gabriel Victor, Stogner, Ronald Earl, Yang, Eric Jian, Zhang, Chaowang "Ricky", Dutta, Partha, Gong, Qing
• Primary Examiner(s): Aspinwall, Evan
• Application Number:
  US16/729,409
• Time in Patent Office: 142 Days
• Field of Search: 707827
• CPC Class Codes: G06F 16/164 File meta data generation
• CPC Class Codes: G06F 16/1727 Details of free space management performed by the file system saving storage space on storage systems G06F3/0608; management of blocks in storage devices G06F3/064
• CPC Class Codes: G06F 16/182 Distributed file systems
• CPC Class Codes: G06F 16/1827 Management specifically adapted to NAS management of storage area networks [SAN] G06F3/067
• CPC Class Codes: G06F 16/9014 hash tables
• CPC Class Codes: G06F 16/90344 by using string matching techniques
• CPC Class Codes: G06F 8/51 Source to source
• CPC Class Codes: G06F 9/46 Multiprogramming arrangements
• CPC Class Codes: G06F 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
• CPC Class Codes: G06N 3/044 Recurrent networks, e.g. Hopfield networks
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 3/084 Backpropagation, e.g. using gradient descent
• CPC Class Codes: H04L 63/102 Entity profiles
• CPC Class Codes: H04L 67/02 based on web technology, e.g. hypertext transfer protocol [HTTP]
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]
• CPC Class Codes: H04L 67/12 specially adapted for proprietary or special-purpose networking environments, e.g. medical networks, sensor networks, networks in vehicles or remote metering networks
• CPC Class Codes: H04L 67/34 involving the movement of software or configuration parameters network booting or remote initial program loading [RIPL] G06F9/4416
• CPC Class Codes: H04L 67/55 : Push-based network services

Abstract

An apparatus includes a processor to: receive a request from a remote device to perform a job flow; retrieve a job flow definition defining the job flow and each of a set of task routines to perform tasks of the job flow from a set of storage devices where each is stored as an undivided object within one storage device; and in response to determining that a data set is stored as multiple data object blocks, generate a container containing the job flow definition and set of task routines to enable each storage device to perform the job flow using a locally stored data object block of the data set as input to generate a corresponding data object block of a result report, provide a copy of the container to each storage device, and transmit the result report assembled from the data object blocks thereof to the remote device.

Claims

1. An apparatus comprising a processor and a storage to store instructions that, when executed by the processor, cause the processor to perform operations comprising:

receive, at the processor, a first request to store a flow input data set in a federated area, wherein;

at least one federated area is defined within storage space provided by at least one of a set of storage devices to store objects to perform a job flow;

the objects to perform the job flow comprise a job flow definition that defines the job flow as a set of tasks to be performed, and a corresponding set of task routines to perform the set of tasks;

processors associated with the set of storage devices cooperate to maintain a distributed file system as spanning storage spaces provided by each storage device of the set of storage devices; and

as part of maintaining the distributed file system, at least one processor associated with of at least one storage device of the set of storage devices determines whether a data object received by the set of storage devices is to be stored as an undivided object or stored as a set of data object blocks into which the received data object is divided and distributed among the set of storage devices based on a size of the received data object compared to a distribution block size;

compare a size of the flow input data set to a threshold size that is based on the distribution block size to determine whether the size of the flow input data set is larger than the threshold size; and

in response to a determination that the size of the flow input data set is larger than the threshold size, the processor is caused to perform operations comprising;

analyze the flow input data set to determine whether the flow input data set is of a distributable form in which data items of the flow input data set are organized into a single homogeneous data structure such that, after the flow input data set is divided into a set of data object blocks, the data items remain accessible from each data object block of the flow input data set independently of the other data object blocks of the flow input data set;

in response to a determination that the flow input data set is not of the distributable form of the flow input data set, the processor is caused to perform operations comprising;

convert the flow input data set from an original form and into the distributable form of the flow input data set; and

following conversion of the original form of the flow input data set into the distributable form, provide the distributable form of the flow input data set to the set of storage devices to be divided by the set of storage devices into the set of data object blocks of the flow input data set that are to be stored in a distributed manner within a first federated area of the at least one federated area, wherein the first federated area is defined within the distributed file system.

2. The apparatus of claim 1, wherein the processor is caused, in response to a determination that the original form of the flow input data set is the distributable form of the flow input data set, to provide the original form of the flow input data set to the set of storage devices to be divided by the set of storage devices into the set of data object blocks of the flow input data set that are to be stored in a distributed manner within the first federated area.

3. The apparatus of claim 1, wherein the processor is caused, in response to a determination that the size of the flow input data set is smaller than the predetermined threshold size, to provide the flow input data set to the set of storage devices to be stored as an undivided object within storage space provided by a single storage device of the set of storage devices, and within a federated area comprising at least one of:

the first federated area defined within the distributed file system; and

a second federated area defined within storage space of a local file system maintained by one storage device of the set of storage devices.

4. The apparatus of claim 1, wherein the processor is caused to perform operations comprising:

retrieve, from the set of storage devices, an indication of the distribution block size; and

define the threshold size to be less than or equal to the distribution block size.

5. The apparatus of claim 1, wherein the processor is caused to perform operations comprising:

receive, at the processor and from a remote device, a second request to retrieve the flow input data set;

retrieve a stored indication of at least the size of the flow input data set;

compare the size of the flow input data set to the threshold size to determine whether the size of the flow input data set is larger than the threshold size;

analyze the retrieved indication to determine whether the flow input data set was converted into the distributable form from the original form prior to being provided to the set of storage devices; and

in response to a determination that the size of the flow input data set is smaller than the threshold size or a determination that the flow input data set was not converted from the original form and into the distributable form, the processor is caused to perform operations comprising;

retrieve the original form of the flow input data set from the one or more storage devices; and

transmit the original form to the remote device.

6. The apparatus of claim 5, wherein, in response to a determination that the size of the flow input data set is larger than the threshold size and a determination that the flow input data set was converted from the original form and into the distributable form, the processor is caused to perform operations comprising:

retrieve a stored indication of one or more characteristics of the original form of the flow input data set;

retrieve the distributable form of the flow input data set from the one or more storage devices;

employ the indication of the one or more characteristics to reverse the conversion to of the flow input data set to re-generate the original form; and

transmit the original form to the remote device.

7. The apparatus of claim 6, wherein the stored indication of one or more characteristics comprises at least one of:

a copy of metadata incorporated into the original form;

an indication of a characteristic of at least one data structure by which data values are organized within the original form; and

an indication of a characteristic of an indexing scheme by which data values are accessed within the original form.

8. The apparatus of claim 1, wherein:

the distributed file system is Hadoop distributed file system (HDFS); and

the distributable form of the flow input data set comprises at least one of;

a text file comprising data items separated by delimiters; and

a optimized row columnar (ORC) file comprising compressed data items.

9. The apparatus of claim 1, wherein the processor is caused to perform operations comprising:

receive, at the processor and from a remote device, a second request to perform the job flow using the flow input data set as an input to the job flow performance, wherein;

at least one result report is to be generated as an output of the job flow performance; and

the job flow definition and each task routine of the set of task routines is stored as an undivided object within one storage device of a set of storage devices;

retrieve the job flow definition and each task routine of the set of task routines from one or more storage devices of the set of storage devices;

retrieve a stored indication of at least the size of the flow input data set;

compare the size of the flow input data set to the threshold size to determine whether the size of the flow input data set is larger than the threshold size; and

in response to the determination that the flow input data set is larger than the threshold size, the processor is caused to perform operations comprising;

generate a container that contains the job flow definition and the set of task routines to enable the processor associated with each storage device to independently perform an instance of the job flow using one of the data object blocks of the flow input data set stored locally within the storage device as an input to the instance, wherein each performance of an instance of the job flow within each storage device generates a corresponding data object block of a set of data object blocks of the result report; and

provide a copy of the container to each storage device of the set of storage devices to enable the processors associated with at least two storage devices of the set of storage devices to perform instances of the job flow at least partially in parallel.

10. The apparatus of claim 9, wherein the processor is caused to perform operations comprising:

retrieve, from each storage device of the multiple storage devices, at least one data object block of the set of data object blocks of the result report;

assemble the result report from the set of data object blocks of the result report;

transmit the result report to the remote device;

compare a size of the result report to the threshold size to determine whether the size of the result report is larger than the threshold size; and

in response to a determination that the size of the result report is larger than the threshold size, the processor is caused to perform operations comprising;

analyze the result report to determine whether the result report is of a distributable form of the result report;

in response to a determination that the result report is not of the distributable form of the result report, convert the result report into the distributable form of the result report; and

provide the distributable form of the result report to the set of storage devices to be divided into the set of data object blocks of the result report that are to be distributed among the set of storage devices.

11. A computer-program product tangibly embodied in a non-transitory machine-readable storage medium, the computer-program product including instructions operable to cause a processor to perform operations comprising:

receive, at the processor, a first request to store a flow input data set in a federated area, wherein;

at least one federated area is defined within storage space provided by at least one of a set of storage devices to store objects to perform a job flow;

the objects to perform the job flow comprise a job flow definition that defines the job flow as a set of tasks to be performed, and a corresponding set of task routines to perform the set of tasks;

processors associated with the set of storage devices cooperate to maintain a distributed file system as spanning storage spaces provided by each storage device of the set of storage devices; and

as part of maintaining the distributed file system, at least one processor associated with of at least one storage device of the set of storage devices determines whether a data object received by the set of storage devices is to be stored as an undivided object or stored as a set of data object blocks into which the received data object is divided and distributed among the set of storage devices based on a size of the received data object compared to a distribution block size;

compare a size of the flow input data set to a threshold size that is based on the distribution block size to determine whether the size of the flow input data set is larger than the threshold size; and

in response to a determination that the size of the flow input data set is larger than the threshold size, the processor is caused to perform operations comprising;

analyze the flow input data set to determine whether the flow input data set is of a distributable form in which data items of the flow input data set are organized into a single homogeneous data structure such that, after the flow input data set is divided into a set of data object blocks, the data items remain accessible from each data object block of the flow input data set independently of the other data object blocks of the flow input data set;

in response to a determination that the flow input data set is not of the distributable form of the flow input data set, the processor is caused to perform operations comprising;

convert the flow input data set from an original form and into the distributable form of the flow input data set; and

following conversion of the original form of the flow input data set into the distributable form, provide the distributable form of the flow input data set to the set of storage devices to be divided by the set of storage devices into the set of data object blocks of the flow input data set that are to be stored in a distributed manner within a first federated area of the at least one federated area, wherein the first federated area is defined within the distributed file system.

12. The computer-program product of claim 11, wherein the processor is caused, in response to a determination that the original form of the flow input data set is the distributable form of the flow input data set, to provide the original form of the flow input data set to the set of storage devices to be divided by the set of storage devices into the set of data object blocks of the flow input data set that are to be stored in a distributed manner within the first federated area.

13. The computer-program product of claim 11, wherein the processor is caused, in response to a determination that the size of the flow input data set is smaller than the predetermined threshold size, to provide the flow input data set to the set of storage devices to be stored as an undivided object within storage space provided by a single storage device of the set of storage devices, and within a federated area comprising at least one of:

the first federated area defined within the distributed file system; and

a second federated area defined within storage space of a local file system maintained by one storage device of the set of storage devices.

14. The computer-program product of claim 11, wherein the processor is caused to perform operations comprising:

retrieve, from the set of storage devices, an indication of the distribution block size; and

define the threshold size to be less than or equal to the distribution block size.

15. The computer-program product of claim 11, wherein the processor is caused to perform operations comprising:

receive, at the processor and from a remote device, a second request to retrieve the flow input data set;

retrieve a stored indication of at least the size of the flow input data set;

compare the size of the flow input data set to the threshold size to determine whether the size of the flow input data set is larger than the threshold size;

analyze the retrieved indication to determine whether the flow input data set was converted into the distributable form from the original form prior to being provided to the set of storage devices; and

in response to a determination that the size of the flow input data set is smaller than the threshold size or a determination that the flow input data set was not converted from the original form and into the distributable form, the processor is caused to perform operations comprising;

retrieve the original form of the flow input data set from the one or more storage devices; and

transmit the original form to the remote device.

16. The computer-program product of claim 15, wherein, in response to a determination that the size of the flow input data set is larger than the threshold size and a determination that the flow input data set was converted from the original form and into the distributable form, the processor is caused to perform operations comprising:

retrieve a stored indication of one or more characteristics of the original form of the flow input data set;

retrieve the distributable form of the flow input data set from the one or more storage devices;

employ the indication of the one or more characteristics to reverse the conversion to of the flow input data set to re-generate the original form; and

transmit the original form to the remote device.

17. The computer-program product of claim 16, wherein the stored indication of one or more characteristics comprises at least one of:

a copy of metadata incorporated into the original form;

an indication of a characteristic of at least one data structure by which data values are organized within the original form; and

an indication of a characteristic of an indexing scheme by which data values are accessed within the original form.

18. The computer-program product of claim 11, wherein:

the distributed file system is Hadoop distributed file system (HDFS); and

the distributable form of the flow input data set comprises at least one of;

a text file comprising data items separated by delimiters; and

a optimized row columnar (ORC) file comprising compressed data items.

19. The computer-program product of claim 11, wherein the processor is caused to perform operations comprising:

receive, at the processor and from a remote device, a second request to perform the job flow using the flow input data set as an input to the job flow performance, wherein;

at least one result report is to be generated as an output of the job flow performance; and

the job flow definition and each task routine of the set of task routines is stored as an undivided object within one storage device of a set of storage devices;

retrieve the job flow definition and each task routine of the set of task routines from one or more storage devices of the set of storage devices;

retrieve a stored indication of at least the size of the flow input data set;

compare the size of the flow input data set to the threshold size to determine whether the size of the flow input data set is larger than the threshold size; and

in response to the determination that the flow input data set is larger than the threshold size, the processor is caused to perform operations comprising;

generate a container that contains the job flow definition and the set of task routines to enable the processor associated with each storage device to independently perform an instance of the job flow using one of the data object blocks of the flow input data set stored locally within the storage device as an input to the instance, wherein each performance of an instance of the job flow within each storage device generates a corresponding data object block of a set of data object blocks of the result report; and

provide a copy of the container to each storage device of the set of storage devices to enable the processors associated with at least two storage devices of the set of storage devices to perform instances of the job flow at least partially in parallel.

20. The computer-program product of claim 19, wherein the processor is caused to perform operations comprising:

retrieve, from each storage device of the multiple storage devices, at least one data object block of the set of data object blocks of the result report;

assemble the result report from the set of data object blocks of the result report;

transmit the result report to the remote device;

compare a size of the result report to the threshold size to determine whether the size of the result report is larger than the threshold size; and

in response to a determination that the size of the result report is larger than the threshold size, the processor is caused to perform operations comprising;

analyze the result report to determine whether the result report is of a distributable form of the result report;

in response to a determination that the result report is not of the distributable form of the result report, convert the result report into the distributable form of the result report; and

provide the distributable form of the result report to the set of storage devices to be divided into the set of data object blocks of the result report that are to be distributed among the set of storage devices.

21. A computer-implemented method comprising:

receiving, by a processor, a first request to store a flow input data set in a federated area, wherein;

at least one federated area is defined within storage space provided by at least one of a set of storage devices to store objects to perform a job flow;

the objects to perform the job flow comprise a job flow definition that defines the job flow as a set of tasks to be performed, and a corresponding set of task routines to perform the set of tasks;

processors associated with the set of storage devices cooperate to maintain a distributed file system as spanning storage spaces provided by each storage device of the set of storage devices; and

as part of maintaining the distributed file system, at least one processor associated with of at least one storage device of the set of storage devices determines whether a data object received by the set of storage devices is to be stored as an undivided object or stored as a set of data object blocks into which the received data object is divided and distributed among the set of storage devices based on a size of the received data object compared to a distribution block size;

comparing, by the processor, a size of the flow input data set to a threshold size that is based on the distribution block size to determine whether the size of the flow input data set is larger than the threshold size; and

in response to a determination that the size of the flow input data set is larger than the threshold size, performing operations comprising;

analyzing, by the processor, the flow input data set to determine whether the flow input data set is of a distributable form in which data items of the flow input data set are organized into a single homogeneous data structure such that, after the flow input data set is divided into a set of data object blocks, the data items remain accessible from each data object block of the flow input data set independently of the other data object blocks of the flow input data set;

in response to a determination that the flow input data set is not of the distributable form of the flow input data set, performing operations comprising;

converting, by the processor, the flow input data set from an original form and into the distributable form of the flow input data set; and

following conversion of the original form of the flow input data set into the distributable form, providing the distributable form of the flow input data set to the set of storage devices to be divided by the set of storage devices into the set of data object blocks of the flow input data set that are to be stored in a distributed manner within a first federated area of the at least one federated area, wherein the first federated area is defined within the distributed file system.

22. The computer-implemented method of claim 21, comprising, in response to a determination that the original form of the flow input data set is the distributable form of the flow input data set, providing the original form of the flow input data set to the set of storage devices to be divided by the set of storage devices into the set of data object blocks of the flow input data set that are to be stored in a distributed manner within the first federated area.

23. The computer-implemented method of claim 21, comprising, in response to a determination that the size of the flow input data set is smaller than the predetermined threshold size, providing the flow input data set to the set of storage devices to be stored as an undivided object within storage space provided by a single storage device of the set of storage devices, and within a federated area comprising at least one of:

the first federated area defined within the distributed file system; and

a second federated area defined within storage space of a local file system maintained by one storage device of the set of storage devices.

24. The computer-implemented method of claim 21, comprising:

retrieving, from the set of storage devices, an indication of the distribution block size; and

defining, by the processor, the threshold size to be less than or equal to the distribution block size.

25. The computer-implemented method of claim 21, comprising:

receiving, by the processor and from a remote device, a second request to retrieve the flow input data set;

retrieving a stored indication of at least the size of the flow input data set;

comparing, by the processor, the size of the flow input data set to the threshold size to determine whether the size of the flow input data set is larger than the threshold size;

analyzing, by the processor, the retrieved indication to determine whether the flow input data set was converted into the distributable form from the original form prior to being provided to the set of storage devices; and

in response to a determination that the size of the flow input data set is smaller than the threshold size or a determination that the flow input data set was not converted from the original form and into the distributable form, performing operations comprising;

retrieving the original form of the flow input data set from the one or more storage devices; and

transmitting, from the processor, the original form to the remote device.

26. The computer-implemented method of claim 25, wherein, in response to a determination that the size of the flow input data set is larger than the threshold size and a determination that the flow input data set was converted from the original form and into the distributable form, performing operations comprising:

retrieving a stored indication of one or more characteristics of the original form of the flow input data set;

retrieving the distributable form of the flow input data set from the one or more storage devices;

employing, by the processor, the indication of the one or more characteristics to reverse the conversion to of the flow input data set to re-generate the original form; and

transmitting, from the processor, the original form to the remote device.

27. The computer-implemented method of claim 26, wherein the stored indication of one or more characteristics comprises at least one of:

a copy of metadata incorporated into the original form;

an indication of a characteristic of at least one data structure by which data values are organized within the original form; and

an indication of a characteristic of an indexing scheme by which data values are accessed within the original form.

28. The computer-implemented method of claim 21, wherein:

the distributed file system is Hadoop distributed file system (HDFS); and

the distributable form of the flow input data set comprises at least one of;

a text file comprising data items separated by delimiters; and

a optimized row columnar (ORC) file comprising compressed data items.

29. The computer-implemented method of claim 21, wherein the processor is caused to perform operations comprising:

receiving, by the processor and from a remote device, a second request to perform the job flow using the flow input data set as an input to the job flow performance, wherein;

at least one result report is to be generated as an output of the job flow performance; and

the job flow definition and each task routine of the set of task routines is stored as an undivided object within one storage device of a set of storage devices;

retrieving the job flow definition and each task routine of the set of task routines from one or more storage devices of the set of storage devices;

retrieving a stored indication of at least the size of the flow input data set;

comparing the size of the flow input data set to the threshold size to determine whether the size of the flow input data set is larger than the threshold size; and

in response to the determination that the flow input data set is larger than the threshold size, the performing operations comprising;

generating, by the processor, a container that contains the job flow definition and the set of task routines to enable the processor associated with each storage device to independently perform an instance of the job flow using one of the data object blocks of the flow input data set stored locally within the storage device as an input to the instance, wherein each performance of an instance of the job flow within each storage device generates a corresponding data object block of a set of data object blocks of the result report; and

providing a copy of the container to each storage device of the set of storage devices to enable the processors associated with at least two storage devices of the set of storage devices to perform instances of the job flow at least partially in parallel.

30. The computer-implemented method of claim 29, comprising:

retrieving, from each storage device of the multiple storage devices, at least one data object block of the set of data object blocks of the result report;

assembling, by the processor, the result report from the set of data object blocks of the result report;

transmitting, from the processor, the result report to the remote device;

comparing, by the processor, a size of the result report to the threshold size to determine whether the size of the result report is larger than the threshold size; and

in response to a determination that the size of the result report is larger than the threshold size, performing operations comprising;

analyzing, by the processor, the result report to determine whether the result report is of a distributable form of the result report;

in response to a determination that the result report is not of the distributable form of the result report, converting, by the processor, the result report into the distributable form of the result report; and

providing the distributable form of the result report to the set of storage devices to be divided into the set of data object blocks of the result report that are to be distributed among the set of storage devices.