Selecting threads for concurrent processing of data

A non-transitory computer readable medium comprising instructions which, when executed by one or more hardware processors, causes performance of operations comprising:

• identifying a task to be completed using multi-threaded processing, the task comprising a plurality of subtasks;
• computing an estimated time for execution of each subtask of the plurality of subtasks;
• selecting a number of threads to execute the plurality of subtasks based on the estimated time for execution of each subtask of the plurality of subtasks;
• and
• executing the plurality of subtasks using the selected number of threads,
• wherein selecting the number of threads comprises:
• computing a first estimated total time to execute the task when using n number of threads, based on the estimated time for execution of each subtask of the plurality of subtasks in the task;
• computing a second estimated total time to execute the plurality of subtasks when using n+m number of threads;
• determining whether the second estimated total time meets a threshold level of improvement over the first estimated total time;
• responsive to determining that the second estimated total time does not meet a threshold level of improvement over the first estimated total time, selecting n number of threads;
• and
• responsive to determining that the second estimated total time meets the threshold level of improvement over the first estimated total time, selecting n+m number of threads.

A non-transitory computer readable medium comprising instructions which, when executed by one or more hardware processors, causes performance of operations comprising:

• identifying a task to be completed using multi-threaded processing, the task comprising a plurality of subtasks;
• computing an estimated time for execution of each subtask of the plurality of subtasks;
• selecting a number of threads to execute the plurality of subtasks based on the estimated time for execution of each subtask of the plurality of subtasks;
• and
• executing the plurality of subtasks using the selected number of threads,
• wherein computing the estimated time for execution of a particular subtask of the plurality of subtasks comprises using a time estimation model, and
• the operations further comprising creating the time estimation model comprising:
• extracting data from a set of previously executed subtasks, the data including feature values for a plurality of features and a completion time for each subtask of the set of previously executed subtasks;
• analyzing the extracted data to identify a set of predictive features that are a subset of features of the set of previously executed subtasks, wherein values of the predictive features for each subtask of the previously executed subtasks are determined to have the greatest impact on an amount of time taken to complete execution of the respective subtask;
• determining relationships between predictive feature values and respective completion time for subtasks;
• and
• generating the time estimation model based on the relationships.

A method comprising:

• identifying a task to be completed using multi-threaded processing, the task comprising a plurality of subtasks;
• computing an estimated time for execution of each subtask of the plurality of subtasks;
• selecting a number of threads to execute the plurality of subtasks based on the estimated time for execution of each subtask of the plurality of subtasks;
• and
• executing the plurality of subtasks using the selected number of threads,
• wherein the method further comprises:
• monitoring a time of execution for each of a second plurality of subtasks;
• determining relationships between features of the second plurality of subtasks and a corresponding time of execution;
• generating a time estimation model based on the relationships;
• determining a set of features corresponding to a particular subtask of the plurality of subtasks;
• and
• applying the set of features to the time estimation model to compute the estimated time for execution of the particular subtasks.

A system comprising:

• at least one hardware device including a processor;
• and
• the system configured to perform operations comprising:
• identifying a task to be completed using multi-threaded processing, the task comprising a plurality of subtasks;
• computing an estimated time for execution of each subtask of the plurality of subtasks;
• selecting a number of threads to execute the plurality of subtasks based on the estimated time for execution of each subtask of the plurality of subtasks;
• and
• executing the plurality of subtasks using the selected number of threads,
• wherein the system is configured to perform further operations comprising:
• monitoring a time of execution for each of a second plurality of subtasks;
• determining relationships between features of the second plurality of subtasks and a corresponding time of execution;
• generating a time estimation model based on the relationships;
• determining a set of features corresponding to a particular subtask of the plurality of subtasks;
• and
• applying the set of features to the time estimation model to compute the estimated time for execution of the particular subtasks.