Technologies for offloading acceleration task scheduling operations to accelerator sleds
Technologies for offloading acceleration task scheduling operations to accelerator sleds include a compute device to receive a request from a compute sled to accelerate the execution of a job, which includes a set of tasks. The compute device is also to analyze the request to generate metadata indicative of the tasks within the job, a type of acceleration associated with each task, and a data dependency between the tasks.
1. One or more non-transitory machine-readable storage media comprising a plurality of instructions stored thereon that, in response to being executed, cause one or more processors to:
execute an orchestrator to:
receive a request to perform a workload, wherein the workload includes a set of tasks;
determine availability data of multiple accelerator devices to perform the workload, wherein the availability data is indicative of which of the one or more of the set of tasks are acceptable for execution on an associated accelerator device of the multiple accelerator devices and which of the one or more of the set of tasks are not acceptable for execution on an associated accelerator device of the multiple accelerator devices;
select one or more of the multiple accelerator devices to perform one or more of the set of tasks based on the availability data and also based on capability to complete the one or more of the set of tasks, wherein the capability to complete the one or more of the set of tasks is based on a type of acceleration for the one or more of the set of tasks, wherein the capability to complete the one or more of the set of tasks is based on capability for parallel execution of the one or more of the set of tasks, and wherein the type of acceleration for the one or more of the set of tasks comprises one or more of: cryptographic operation or data compression; and
cause execution of the one or more of the set of tasks on the selected one or more of the multiple accelerator devices.
2. The one or more non-transitory machine-readable storage media of claim 1 , wherein the capability to complete the one or more of the set of tasks is based on data dependence between the one or more of the set of tasks.
3. The one or more non-transitory machine-readable storage media of claim 1 , wherein the capability to complete the one or more of the set of tasks is based on subdivision of the one or more of the set of tasks to operate on different portions of a data set concurrently.
4. The one or more non-transitory machine-readable storage media of claim 1 , wherein the capability to complete the one or more of the set of tasks is based on an estimated time to complete the one or more of the set of tasks.
5. The one or more non-transitory machine-readable storage media of claim 1 , wherein the capability to complete the one or more of the set of tasks is based on network congestion.
6. The one or more non-transitory machine-readable storage media of claim 1 , wherein the capability to complete the one or more of the set of tasks is based on an earliest received indication of capability to complete the one or more of the set of tasks.
7. The one or more non-transitory machine-readable storage media of claim 1 , wherein the one or more accelerator devices comprise one or more of: a processor, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or a graphics processing unit (GPU).
8. An apparatus comprising:
an input/output (I/O) subsystem and
circuitry to:
receive a request to perform a workload, wherein the workload includes a set of tasks;
determine availability data of multiple accelerator devices to perform the workload, wherein the availability data is indicative of which of the one or more of the set of tasks are acceptable for execution on an associated accelerator device of the multiple accelerator devices and which of the one or more of the set of tasks are not acceptable for execution on an associated accelerator device of the multiple accelerator devices;
select one or more of the multiple accelerator devices to perform one or more of the set of tasks based on the availability data and also based on capability to complete the one or more of the set of tasks, wherein the capability to complete the one or more of the set of tasks is based on a type of acceleration for the one or more of the set of tasks, wherein the capability to complete the one or more of the set of tasks is based on capability for parallel execution of the one or more of the set of tasks, and wherein the type of acceleration for the one or more of the set of tasks comprises one or more of: cryptographic operation or data compression; and
cause execution of the one or more of the set of tasks on the selected one or more of the multiple accelerator devices.
9. The apparatus of claim 8 , wherein the capability to complete the one or more of the set of tasks is based on data dependence between the one or more of the set of tasks.
10. The apparatus of claim 8 , wherein the capability to complete the one or more of the set of tasks is based on subdivision of the one or more of the set of tasks to operate on different portions of a data set concurrently.
11. The apparatus of claim 8 , wherein the capability to complete the one or more of the set of tasks is based on an estimated time to complete the one or more of the set of tasks.
12. The apparatus of claim 8 , comprising the one or more accelerator devices coupled to the I/O subsystem.
13. A method comprising:
receiving a request to perform a workload, wherein the workload includes a set of tasks;
determining availability data of multiple accelerator devices to perform the workload, wherein the availability data is indicative of which of the one or more of the set of tasks are acceptable for execution on an associated accelerator device of the multiple accelerator devices and which of the one or more of the set of tasks are not acceptable for execution on an associated accelerator device of the multiple accelerator devices;
selecting one or more of the multiple accelerator devices to perform one or more of the set of tasks based on the availability data and also based on capability to complete the one or more of the set of tasks, wherein the capability to complete the one or more of the set of tasks is based on a type of acceleration for the one or more of the set of tasks, wherein the capability to complete the one or more of the set of tasks is based on capability for parallel execution of the one or more of the set of tasks, and wherein the type of acceleration for the one or more of the set of tasks comprises one or more of: cryptographic operation or data compression; and
causing execution of the one or more of the set of tasks on the selected one or more of the multiple accelerator devices.
14. The method of claim 13 , wherein the capability to complete the one or more of the set of tasks is based on subdivision of the one or more of the set of tasks to operate on different portions of a data set concurrently.
15. The method of claim 13 , wherein the capability to complete the one or more of the set of tasks is based on an estimated time to complete the one or more of the set of tasks.