US 12,730,719 B2Granted

Systems and methods for optimizing job execution in on-demand code execution system

AssigneeFmr LLC

InventorsPraveen Narayanasamy, Rahul Narayanan, Dhinu Prashanth Radhakrishnan, Sukanya Kumaraswamy, Jamunadevi Vijayan

Life of this patent

Estimating the term…

Loading patent…

Abstract

Systems and methods are described for optimizing job execution on an on-demand code execution system. More specifically, the job execution manager may receive requests to execute jobs on an on-demand code execution system. The job execution manager may forecast the runtime of the job by using two or more machine learning models. The two or more machine learning models may generate a runtime prediction set that includes two or more runtime predictions based on metadata of the job. A final runtime prediction is determined based on the runtime prediction set. Based on the final runtime prediction, the job execution manager determines a compute instance on which to execute the job. In case that the job execution manager determines an error has occurred in the compute instance, the job execution manager attempts to execute the job on another larger compute instance.

Claims24

Dependent claims

A computing system for optimizing job execution on an on-demand code execution system, the system comprising memory for storing computer-executable instructions and a processor that executes the computer-executable instructions to:

receive instructions to execute a job having user code on the on-demand code execution system which comprises one or more compute instances that are ordered sequentially based on size;
generate, via two or more machine learning models, a runtime prediction set for the job, the runtime prediction set including a runtime prediction that is generated by each machine learning model of the two or more machine learning models, wherein each runtime prediction is a forecast of the runtime associated with executing the job;
assign a computer instance of the one or more compute instances to the job based on a runtime type of the job, wherein the runtime type of the job is determined based on the runtime prediction having the longest runtime;
and
continuously attempt to execute the job on the compute instance, wherein the job is assigned a new compute instance that is larger in size than a current compute instance upon detection of an execution error when attempting to execute the job on the current compute instance.

2 more elements

A computerized method for optimizing job execution on an on-demand code execution system, the method comprising:

receiving instructions to execute a job having user code on the on-demand code execution system which comprises one or more compute instances that are ordered sequentially based on size;
generating, via two or more machine learning models, a runtime prediction set for the job, the runtime prediction set including a runtime prediction that is generated by each machine learning model of the two or more machine learning models, wherein each runtime prediction is a forecast of the runtime associated with executing the job;
assigning a compute instance of the one or more compute instances to the job based on a runtime type of the job, wherein the runtime type of the job is determined based on the runtime prediction having the longest runtime;
and
continuously attempting to execute the job on the compute instance, wherein the job is assigned a new compute instance that is larger in size than a current compute instance upon detection of an execution error when attempting to execute the job on the current compute instance.

2 more elements

Dependencies are inferred from the claim text.