Machine learning based runtime optimization
A method of operating a shared computing environment includes receiving a client request for an optimized runtime configuration of the shared computing environment for executing a task, receiving runtime data of the task collected from a prior runtime, inferring a runtime configuration from the runtime data, determining a predicted timeseries workload for the shared computing environment, updating the runtime configuration to including the timeseries workload prediction, wherein the updating outputs the optimized runtime configuration, and configuring the shared computing environment according to the optimized runtime configuration in executing the task.
1. A method of operating a shared computing environment, the method comprising:
receiving a client request for an optimized runtime configuration of the shared computing environment for executing a task;
receiving runtime data of the task collected from a prior runtime;
inferring a runtime configuration from the runtime data;
determining a predicted timeseries workload for the shared computing environment;
updating the runtime configuration to include the timeseries workload prediction, wherein the updating outputs the optimized runtime configuration; and
configuring the shared computing environment according to the optimized runtime configuration in executing the task.
2. The method of claim 1 , further comprising outputting the optimized runtime configuration to a configuration file.
3. The method of claim 1 , wherein inferring the runtime configuration further comprises identifying a pattern template for the task based on the runtime data.
4. The method of claim 3 , wherein identifying the pattern template further comprises:
determining a use case profile for the task to discover a pattern of the task; and
performing a pattern analysis using the pattern discovered for the task to identify the pattern template.
5. The method of claim 4 , wherein the pattern analysis comprises at least one of a label matching, a clustering analysis, and a similarity analysis.
6. The method of claim 4 , wherein identifying the pattern template further comprises:
determining a confidence for the pattern template determined by the pattern analysis; and
identifying the pattern template determined by the pattern analysis as an identified pattern upon determining that the confidence is greater than a threshold.
7. The method of claim 4 , wherein identifying the pattern template further comprises:
determining a confidence for the pattern template determined by the pattern analysis; and
identifying the pattern template determined by the pattern analysis as a general pattern upon determining that the confidence is less than a threshold.
8. The method of claim 1 , wherein the optimized runtime configuration deploys limited processor and memory resources of the shared computing environment to perform the task based on the timeseries workload prediction, and wherein the task is a model training.
9. A non-transitory computer readable storage medium comprising computer executable instructions which when executed by a computer cause the computer to perform a method of operating a shared computing environment, the method comprising:
receiving a client request for an optimized runtime configuration of the shared computing environment for executing a task;
receiving runtime data of the task collected from a prior runtime;
inferring a runtime configuration from the runtime data;
determining a predicted timeseries workload for the shared computing environment;
updating the runtime configuration to include the timeseries workload prediction, wherein the updating outputs the optimized runtime configuration; and
configuring the shared computing environment according to the optimized runtime configuration in executing the task.
10. The non-transitory computer readable storage medium of claim 9 , further comprising outputting the optimized runtime configuration to a configuration file.
11. The non-transitory computer readable storage medium of claim 9 , wherein inferring the runtime configuration further comprises identifying a pattern template for the task based on the runtime data.
12. The non-transitory computer readable storage medium of claim 11 , wherein identifying the pattern template further comprises:
determining a use case profile for the task to discover a pattern of the task; and
performing a pattern analysis using the pattern discovered for the task to identify the pattern template.
13. The non-transitory computer readable storage medium of claim 12 , wherein the pattern analysis comprises at least one of a label matching, a clustering analysis, and a similarity analysis.
14. The non-transitory computer readable storage medium of claim 12 , wherein identifying the pattern template further comprises:
determining a confidence for the pattern template determined by the pattern analysis; and
identifying the pattern template determined by the pattern analysis as an identified pattern upon determining that the confidence is greater than a threshold.
15. The non-transitory computer readable storage medium of claim 12 , wherein identifying the pattern template further comprises:
determining a confidence for the pattern template determined by the pattern analysis; and
identifying the pattern template determined by the pattern analysis as a general pattern upon determining that the confidence is less than a threshold.
16. The non-transitory computer readable storage medium of claim 9 , wherein the optimized runtime configuration deploys limited processor and memory resources of the shared computing environment to perform the task based on the timeseries workload prediction, and wherein the task is a model training.
17. A method of allocating resources of a shared computing environment to a plurality of tasks, the method comprising:
receiving a plurality of client requests for an optimized runtime configuration of the shared computing environment for executing respective tasks;
receiving runtime data of the tasks collected from a prior runtime;
inferring a runtime configuration of each of the tasks from the runtime data;
determining a predicted timeseries workload for the shared computing environment;
updating the runtime configuration of each of the tasks to including the timeseries workload prediction, wherein the updating outputs the optimized runtime configuration of each of the tasks;
scheduling an execution of each of the tasks over a time window according to the optimized runtime configurations of each of the tasks; and
configuring the shared computing environment over the time window according to the optimized runtime configurations in executing the tasks.
18. The method of claim 17 , further comprising outputting the optimized runtime configurations to a configuration file available to a scheduler performing the scheduling.
19. The method of claim 17 , wherein inferring the runtime configurations further comprises identifying a pattern template for each of the tasks based on the runtime data.
20. The method of claim 17 , wherein each of the optimized runtime configurations deploy limited processor and memory resources of the shared computing environment to perform the respective task based on the timeseries workload prediction, and wherein the tasks are model training tasks.