Prevention of resource starvation across stages and/or pipelines in computer environments
A computer-implemented method, in accordance with one aspect of the present invention, includes analyzing timing data for stages in a pipeline running in an edge system for detecting starvation of one or more of the stages in the pipeline. In response to detecting one or more of the stages being starved, starvation avoidance is performed for mitigating the starvation of the starving stage(s). A computer-implemented method, in accordance with another aspect of the present invention, includes analyzing timing data for pipelines running in parallel in an edge system for detecting starvation of one or more of the pipelines. In response to detecting one or more of the pipelines being starved, starvation avoidance is performed for mitigating the starvation of the starving pipeline(s).
1 . A computer-implemented method, comprising:
receiving a workload request for executing a workload in a pipeline running in an edge system;
executing the workload in a plurality of stages of the pipeline according to a level of quality, wherein each stage of the plurality of stages is associated with respective timing data; and
in response to one or more of the stages of the pipeline being starved during the executing based on the respective timing data, mitigating the starvation of the one or more starving stages, wherein the mitigating comprises adapting the workload request to reduce resource consumption of a dominating stage of the plurality of stages by reducing the level of quality of output of the dominating stage and providing resources to at least one of the one or more starving stages, and wherein the dominating stage consumes the most resources relative to the other stages of the pipeline during the executing.
2 . The computer-implemented method of claim 1 , wherein the timing data includes lag times between the plurality of stages associated with a trend of change in one or more of said lag times.
3 . The computer-implemented method of claim 2 , further comprising:
responsive to an increase in a lag time between a first stage and a second stage positioned consecutively in the pipeline, increasing resource usage of the second stage.
4 . The computer-implemented method of claim 2 , further comprising:
responsive to a decrease in a lag time between a first stage and a second stage positioned consecutively in the pipeline, decreasing resource usage of the second stage.
5 . The computer-implemented method of claim 1 , wherein a change in one of a plurality of lag ratios over time is indicative of the starvation, and wherein the plurality of lag ratios is based on lag times between different sets of stages of the pipeline.
6 . The computer-implemented method of claim 5 , further comprising:
responsive to an increase in a lag ratio corresponding to a first stage and a second stage positioned consecutively in the pipeline, reducing resource usage of the first stage and/or increasing resource usage of the second stage.
7 . The computer-implemented method of claim 5 , further comprising:
responsive to a decrease in a lag ratio corresponding to a first stage and a second stage positioned consecutively in the pipeline, increasing resource usage of the first stage and/or decreasing resource usage of the second stage.
8 . The computer-implemented method of claim 5 , further comprising:
adjusting a performance of at least one stage corresponding to a log ratio of a plurality of log ratios with a largest change by instructing a horizontal Pod autoscaler to scale at least one Pod allocated to the at least one stage.
9 . The computer-implemented method of claim 1 , wherein mitigating the starvation further includes scaling at least one Pod operating in one or more of the stages.
10 . The computer-implemented method of claim 1 , wherein mitigating the starvation further includes temporarily suspending operation of at least one of the stages.
11 . The computer-implemented method of claim 1 , wherein mitigating the starvation further includes holding back a workload request from processing by the pipeline up to a threshold amount of time.
12 . The computer-implemented method of claim 1 , wherein mitigating the starvation further includes employing a knowledge base artificial intelligence model.
13 . A system comprising:
a processor; and
one or more computer readable storage media comprising program instructions collectively stored on the one or more computer readable storage media, the program instructions when executed by the processor execute a method comprising:
receiving a workload request for executing a workload in a pipeline running in an edge system;
executing the workload in a plurality of stages of the pipeline according to a level of quality, wherein each stage of the plurality of stages is associated with respective timing data; and
in response to one or more of the stages of the pipeline being starved during the executing based on the respective timing data, mitigating the starvation of the one or more starving stages, wherein the mitigating comprises adapting the workload request to reduce resource consumption of a dominating stage of the plurality of stages by reducing the level of quality of output of the dominating stage and providing resources to at least one of the one or more starving stages, and wherein the dominating stage consumes the most resources relative to the other stages of the pipeline during the executing.
14 . The system of claim 13 , wherein the timing data includes lag times between the plurality of stages associated with a trend of change in one or more of said lag times.
15 . The system of claim 14 , wherein the method further comprises:
responsive to an increase in a lag time between a first stage and a second stage positioned consecutively in the pipeline, increasing resource usage of the second stage.
16 . The system of claim 13 , wherein a change in one of a plurality of lag ratios over time is indicative of the starvation, and wherein the plurality of lag ratios is based on lag times between different sets of stages of the pipeline.
17 . The system of claim 16 , wherein the method further comprises:
responsive to an increase in a lag ratio corresponding to a first stage and a second stage positioned consecutively in the pipeline, reducing resource usage of the first stage and/or increasing resource usage of the second stage.
18 . The system of claim 16 , wherein the method further comprises:
adjusting a performance of at least one stage corresponding to a log ratio of a plurality of log ratios with a largest change by instructing a horizontal Pod autoscaler to scale at least one Pod allocated to the at least one stage.
19 . The system of claim 13 , wherein mitigating the starvation further includes scaling at least one Pod operating in one or more of the stages.
20 . A computer program product, the computer program product comprising:
one or more computer readable storage media and program instructions collectively stored on the one or more computer readable storage media, the program instructions comprising program instructions to:
receive a workload request for executing a workload in a pipeline running in an edge system;
execute the workload in a plurality of stages of the pipeline according to a level of quality, wherein each stage of the plurality of stages is associated with respective timing data; and
in response to one or more of the stages of the pipeline being starved during the execution based on the respective timing data, mitigate the starvation of the one or more starving stages, wherein the mitigating comprises adapting the workload request to reduce resource consumption of a dominating stage of the plurality of stages by reducing the level of quality of output of the dominating stage and providing resources to at least one of the one or more starving stages, and wherein the dominating stage consumes the most resources relative to the other stages of the pipeline during the execution.