Systems and methods for estimating computation times a-priori in fog computing robotics
In order to make use of computational resources available at runtime through fog networked robotics paradigm, it is critical to estimate average performance capacities of deployment hardware that is generally heterogeneous. It is also not feasible to replicate runtime deployment framework, collected sensor data and realistic offloading conditions for robotic environments. In accordance with an embodiment of the present disclosure, computational algorithms are dynamically profiled on a development testbed, combined with benchmarking techniques to estimate compute times over the deployment hardware. Estimation in accordance with the present disclosure is based both on Gustafson's law as well as embedded processor benchmarks. Systems and methods of the present disclosure realistically capture parallel processing, cache capacities and differing processing times across hardware.
1. A processor implemented method ( 200 ) comprising:
dynamically profiling computational tasks on a development testbed for a-priori estimation of computational time and energy requirements for executing the computational tasks based on a plurality of hardware performance counters ( 202 ); and
extrapolating the computational time and the energy requirements for executing the computational tasks on one or more deployment hardware based on benchmarks and parallel processing models and further based on number of cores and rated frequency of the Central Processing Unit (CPU) associated with the one or more deployment hardware ( 204 );
dynamically allocating the computational tasks to the one or more deployment hardware based on the extrapolated computational time and the energy requirements by applying pre-defined rules.
2. The processor implemented method of claim 1 , wherein the one or more deployment hardware is heterogeneous and comprises one or more of robot nodes, fog nodes and cloud virtual machine nodes.
3. The processor implemented method of claim 1 , wherein the pre-defined rules are based on computational complexity, latency constraints and processing power associated with the one or more deployment hardware.
4. A system ( 100 ) comprising:
one or more data storage devices ( 102 ) operatively coupled to one or more hardware processors ( 104 ) and configured to store instructions configured for execution by the one or more hardware processors to:
dynamically profile computational tasks on a development testbed for a-priori estimation of computational time and energy requirements for executing the computational tasks based on a plurality of hardware performance counters; and
extrapolate the computational time and the energy requirements for executing the computational tasks on one or more deployment hardware based on benchmarks and parallel processing models and further based on number of cores and rated frequency of the Central Processing Unit (CPU) associated with the one or more deployment hardware;
intelligently allocate the computational tasks to the one or more deployment hardware based on the extrapolated computational time and the energy requirements by applying pre-defined rules.
5. The system of claim 4 , wherein the one or more deployment hardware is heterogeneous and comprises one or more of robot nodes, fog nodes and cloud virtual machine nodes.
6. The system of claim 4 , wherein the pre-defined rules are based on computational complexity, latency constraints and processing power associated with the one or more deployment hardware.
7. A computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:
dynamically profiling computational tasks on a development testbed for a-priori estimation of computational time and energy requirements for executing the computational tasks based on a plurality of hardware performance counters; and
extrapolating the computational time and the energy requirements for executing the computational tasks on one or more deployment hardware based on benchmarks and parallel processing models and further based on number of cores and rated frequency of the Central Processing Unit (CPU) associated with the one or more deployment hardware;
intelligently allocating the computational tasks to the one or more deployment hardware based on the extrapolated computational time and the energy requirements by applying pre-defined rules.