Approximate computing for application performance in heterogeneous systems
A system and method for determining a configuration of a plurality of tasks to meet the specified deadline of a linear workflow of a real-time heterogeneous network. Often times, while meeting expected application performance in the heterogeneous network, it may possible to have graceful degradation of quality for ensuring timing constraints at the same time. In a multi-layered architecture, where each layer is equipped with multiple computational resources, the time optimization for each of the plurality of tasks can be achieved through approximate computing and analyzing all possible configurations of each task in a workflow within a particular layer.
1. A method for determining a configuration of a plurality of tasks to meet a specified deadline of a linear workflow of a real-time heterogeneous network, the method comprising:
selecting, by a processor, a task from the plurality of tasks in the linear workflow of the real-time heterogeneous network, wherein the real-time heterogeneous network has a multi-layered architecture, and wherein each task of the plurality of tasks has different realizations with different execution times on different layers of the heterogeneous network;
considering, by the processor, a realization from the plurality of realizations of the selected task, wherein the plurality of realizations approximate errors of the selected task and a relative quality among the plurality of realizations of the selected task, and wherein the relative quality is based on one or more orders of operations performed in the realization to meet the specified deadline;
defining, by the processor, a realization index based on the considered realization of the selected task of the linear workflow;
analyzing, by the processor, a plurality of configurations for the selected task, wherein each of the plurality of configurations depends on a plurality of implementations and a plurality of physical parts of the heterogeneous network;
estimating, by the processor, execution time for each of the analyzed plurality of configurations, wherein the estimation follows a recursive routine and completes the execution within the specified deadline on a cost of quality of output, and wherein the processor optimizes the execution time based on measured values received from a profiler monitoring the plurality of tasks in the linear workflow; and
determining, by the processor, a configuration for the selected task based on the least execution time estimation for each of the plurality of tasks to meet the specified deadlines of the workflow of a real-time heterogeneous network.
2. The method claimed in claim 1 , wherein one or more resources within a layer of the heterogeneous network are homogeneous.
3. The method claimed in claim 1 , wherein each of the plurality of realizations depends on one or more limitations of a layer of the heterogeneous network.
4. The method claimed in claim 3 , wherein the one or more limitations of the layer includes execution time, power consumption values and quality of output.
5. A system for determining a task configuration to meet specified deadlines of a plurality of tasks of a real-time heterogeneous network, the system comprising:
a processor;
a memory communicatively coupled to the processor and the memory contains instructions that are readable by the processor and, when executed by the processor, cause the processor to:
select a task from the plurality of tasks in a linear workflow of the real-time heterogeneous network, wherein the real-time heterogeneous network has a multi-layered architecture, and wherein each task of the plurality of tasks has different realizations with different execution times on different layers of the heterogeneous network;
consider a realization from the plurality of realizations of the selected task, wherein the plurality of realizations approximate errors of the selected task and a relative quality among the plurality of realizations of the selected task, and wherein the relative quality is based on one or more orders of operations performed in the realization to meet the specified deadline;
analyze a plurality of configurations for the selected task, wherein each of the plurality of configurations depends on a plurality of implementations and a plurality of physical parts of the heterogeneous network;
estimate execution time for each of the analyzed plurality of configurations, wherein the estimation follows a recursive routine and completes the execution within the specified deadline on a cost of quality of output, and wherein the processor optimizes the execution time based on measured values received from a profiler monitoring the plurality of tasks in the linear workflow; and
determine a configuration for the selected task based on the least execution time estimation for each of the plurality of tasks to meet the specified deadlines of the workflow of a real-time heterogeneous network.
6. The system claimed in claim 5 , wherein one or more resources within a layer of the heterogeneous network are homogeneous.
7. The system claimed in claim 5 , wherein each of the plurality of realizations depends on one or more limitations of a layer of the heterogeneous network.
8. The system claimed in claim 7 , wherein the one or more limitations of the layer includes execution time, power consumption values and quality of output.
9. A non-transitory computer readable medium storing instructions, which when executed by one or more hardware processors, causes the one or more hardware processors to execute a method for determining a configuration of a plurality of tasks to meet a specified deadline of a linear workflow of a real-time heterogeneous network, the method comprising:
selecting a task from the plurality of tasks in the linear workflow of the real-time heterogeneous network, wherein the real-time heterogeneous network has a multi-layered architecture, and wherein each task of the plurality of tasks has different realizations with different execution time on different layers of the heterogeneous network;
considering a realization from the plurality of realizations of the selected task, wherein the plurality of realizations approximate errors of the selected task and a relative quality among the plurality of realizations of the selected task, and wherein the relative quality is based on one or more orders of operations performed in the realization to meet the specified deadline;
defining a realization index based on the considered realization of the selected task of the linear workflow;
analyzing a plurality of configurations for the selected task, wherein each of the plurality of configurations depends on a plurality of implementations and a plurality of physical parts of the heterogeneous network;
estimating execution time for each of the analyzed plurality of configurations, wherein the estimation follows a recursive routine and completes the execution within the specified deadline on a cost of quality of output, and wherein the one or more hardware processors optimize the execution time based on measured values received from a profiler monitoring the plurality of tasks in the linear workflow; and
determining a configuration for the selected task based on the least execution time estimation for each of the plurality of tasks to meet the specified deadlines of the workflow of a real-time heterogeneous network.