Full exploitation of parallel processors for data processing

• US 8,959,524 B2
• Filed: 06/28/2012
• Issued: 02/17/2015
• Est. Priority Date: 01/30/2012
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Hirsch, Michael, Klein, Shmuel Thomas, Toaff, Yair
• Primary Examiner(s): Bullock, Jr., Lewis A
• Assistant Examiner(s): ALFRED, MELISSA A
• Application Number:
  US13/536,320
• Publication Number:
  US 20130198754A1
• Time in Patent Office: 964 Days
• Field of Search: None
• US Class Current: 718/104
• CPC Class Codes: G06F 16/215 Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
• CPC Class Codes: G06F 16/2272 Management thereof
• CPC Class Codes: G06F 2209/485 Resource constraint
• CPC Class Codes: G06F 2209/5012 Processor sets
• CPC Class Codes: G06F 2209/506 Constraint
• CPC Class Codes: G06F 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
• CPC Class Codes: G06F 9/5066 Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs mappping at compile time, see G06F8/451
• CPC Class Codes: G06F 9/5077 Logical partitioning of resources; Management or configuration of virtualized resources specific details on emulation or internal functioning of virtual machines G06F9/455
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Exemplary method, system, and computer program product embodiments for full exploitation of parallel processors for data processing are provided. In one embodiment, by way of example only, a set of parallel processors is partitioned into disjoint subsets according to indices of the set of the parallel processors. The size of each of the disjoint subsets corresponds to a number of processors assigned to the processing of the data chunks at one of the layers. Each of the processors are assigned to different layers in different data chunks such that each of processors are busy and the data chunks are fully processed within a number of the time steps equal to the number of the layers. A transition function is devised from the indices of the set of the parallel processors at one time steps to the indices of the set of the parallel processors at a following time step.

Claims

1. A method for full exploitation, after a plurality of initialization steps, of a set of a plurality of parallel processors to perform a task on a sequence of data chunks by a processor device in a computing environment, wherein each of the data chunks are processed in several time steps and by a plurality of layers with the plurality of layers being dealt with by at least one of a plurality of processors at each of the time steps, the method comprising:

partitioning the set of the plurality of parallel processors into disjoint subsets according to indices of the set of the plurality of parallel processors wherein,

the plurality of parallel processors are partitioned in accordance with one or more of a plurality of constraints;

the number of the plurality of parallel processors that are available is n=2^d−

1, where d is the number of the plurality of layers, n is the number of the plurality of parallel processors, and 2^d−

1 is also equal to n=(2^d)−

1, and

a size of each of the disjoint subsets corresponds to a number of the plurality of processors assigned to the processing of the data chunks at one of the plurality of layers;

partitioning the task into the plurality of layers independent of partitioning the set of the plurality of parallel processors;

assigning each of the plurality of processors to the plurality of layers of the task according to the partitioning of the task such that each of the plurality of processors are busy and each of the data chunks are fully processed within a number of the time steps equal to the number of the plurality of layers, wherein the number of the plurality of parallel processors assigned to the processing of the data chunks at one of the plurality of layers is smaller than the number of the plurality of parallel processors assigned to the processing of the data chunks at a preceding one of the plurality of layers;

selecting and using one of a plurality of constraints for restricting any one of the plurality of parallel processors to always work on the same one of the plurality of layers or, for each of the plurality of layers except for a first layer, to always work on a same data chunk as in each of the previous layers;

processing each one of the data chunks within a number of time steps equal to a number of the plurality of layers; and

devising a transition function from the indices of the set of the plurality of parallel processors at one of the time steps to the indices of the set of the plurality of parallel processors at a following time step, wherein the transition function is consistent with the assigning.

2. The method of claim 1, further including, subjecting the transition function to the plurality of constraints.

3. The method of claim 2, wherein the plurality of constraints include restricting at least one of the plurality of processors at each of the time steps to the same one of the plurality of layers that is assigned for the processing of the data chunks.

4. The method of claim 2, wherein the plurality of constraints include restricting the set of the plurality of parallel processors that is assigned to the processing of the data chunks at one of the plurality of layers which is not the first to be a subset of the set of the plurality of parallel processors assigned to the processing of the data chunks at previous one of the plurality of layers.

5. The method of claim 1, wherein

the number of the plurality of processors assigned to level 0 is n=2^d−

1, and

the number of the plurality of processors assigned to the processing of the data chunks at one of the plurality of layers which is not the first is half of the number of the plurality of processors assigned to the processing of the data chunks at the previous one of the plurality of layers.

6. The method of claim 5, wherein, if at least one of the plurality of processors, at each of the time steps, is restricted to being assigned to the same one of the plurality of layers for processing of the data chunks:

the plurality of parallel processors are indexed sets by the integers 0 to 2n−

2,

the number of the plurality of processors assigned to the processing of the data chunks at one of the plurality of layers indexed i, for i=0, 1, . . . , d−

1 is n/2ⁱ, and

the indices of the plurality of parallel processors of n/2ⁱ are n/2ⁱ numbers that have a standard binary representation of length d+1 ending in 0 followed by i ones.

7. The method of claim 5, further including, for each of the plurality of layers except a first one of the plurality of layers, restricting each one of the plurality of processors to the processing of a same data chunk of a previous layers of the plurality of layers, wherein:

the plurality of parallel processors are indexed sets by the integers 1 to 2n−

1,

the number of the plurality of processors assigned to the processing of the data chunks at one of the plurality of layers indexed i, for i=0, 1, . . . , d−

1 is n/2ⁱ, and

the indices of the plurality of parallel processors of n/2ⁱ are n/2ⁱ numbers that have a standard binary representation of length d+1 ending in 0 followed by i zeros.

8. The method of claim 1, further including, in conjunction with the devising, choosing the transition function as applying a cyclical shift by at least one bit to the left to a standard binary representation of the indices of the plurality of processors.