Technologies for coordinating disaggregated accelerator device resources

A method implemented using a cloud computing system, the cloud computing system comprising physical cloud computing resources, the physical cloud computing resources being configurable to comprise physical compute resources, physical accelerator resources, and management resources, the physical compute resources including at least one central processing unit (CPU) and memory circuitry, the physical accelerator resources including graphics processing unit (GPU) accelerator circuitry, the method comprising:

• scheduling, by the management resources, multiple portions of at least one workload to at least one set of multiple virtualized resources for concurrent execution by the at least one set of the multiple virtualized resources, the at least one workload comprising at least one virtual machine and/or at least one container, the physical compute resources and the physical accelerator resources to be used to provide the multiple virtualized resources;
• and
• dynamically reallocating, by the management resources, one or more of the multiple portions of the at least one workload to be concurrently executed by at least one other set of the multiple virtualized resources, the at least one set of the multiple virtualized resources being different, at least in part, from the at least one other set of the multiple virtualized resources;
• wherein:
• the scheduling and/or the dynamically reallocating are to be based, at least in part, upon workload type data associated with the at least one workload, accelerator utilization data, and workload request data;
• and
• the physical compute resources and the physical accelerator resources are configurable to be comprised in multiple local and remote hosts.

A cloud computing system comprising:

• physical cloud computing resources configurable to comprise physical compute resources, physical accelerator resources, and management resources, the physical compute resources comprising at least one central processing unit (CPU) and memory circuitry, the physical accelerator resources including graphics processing unit (GPU) accelerator circuitry;
• wherein:
• the management resources are to schedule multiple portions of at least one workload to at least one set of multiple virtualized resources for concurrent execution by the at least one set of the multiple virtualized resources;
• the at least one workload comprises at least one virtual machine and/or at least one container;
• the physical compute resources and the physical accelerator resources are to be used to provide the multiple virtualized resources;
• the management resources are to dynamically reallocate one or more of the multiple portions of the at least one workload to be concurrently executed by at least one other set of the multiple virtualized resources;
• the at least one set of the multiple virtualized resources is different, at least in part, from the at least one other set of the multiple virtualized resources;
• scheduling, by the management resources, of the multiple portions of the at least one workload and/or dynamically reallocating, by the management resources, of the one or more of the multiple portions of the at least one workload are to be based, at least in part, upon workload type data associated with the at least one workload, accelerator utilization data, and workload request data;
• and
• the physical compute resources and the physical accelerator resources are configurable to be comprised in multiple local and remote hosts.

At least one data center system for use in association with implementing a cloud computing system, the at least one data center system comprising:

• physical cloud computing resources configurable to comprise physical compute resources, physical accelerator resources, and management resources, the physical compute resources comprising at least one central processing unit (CPU) and memory circuitry, the physical accelerator resources including graphics processing unit (GPU) accelerator circuitry;
• wherein:
• the management resources are to schedule multiple portions of at least one workload to at least one set of multiple virtualized resources for concurrent execution by the at least one set of the multiple virtualized resources;
• the at least one workload comprises at least one virtual machine and/or at least one container;
• the physical compute resources and the physical accelerator resources are to be used to provide the multiple virtualized resources;
• the management resources are to dynamically reallocate one or more of the multiple portions of the at least one workload to be concurrently executed by at least one other set of the multiple virtualized resources;
• the at least one set of the multiple virtualized resources are different, at least in part, from the at least one other set of the multiple virtualized resources;
• scheduling, by the management resources, of the multiple portions of the at least one workload and/or dynamically reallocating, by the management resources, of the one or more of the multiple portions of the at least one workload are to be based, at least in part, upon workload type data associated with the at least one workload, accelerator utilization data, and workload request data;
• and
• the physical compute resources and the physical accelerator resources are configurable to be comprised in multiple local and remote hosts.