Abstract: A method for distributed computing between a host computer and at least one accelerator device interconnected through a network includes profiling a data transfer rate and a computation rate for a range of data sizes to find an optimal chunk size for the data transfer through the network; splitting or aggregating a size of the data stored in a memory in the host computer for encapsulating the data into a chunk with the optimal chunk size; dispatching the encapsulated data to the accelerator device; and instructing pipeline computation to the accelerator device with respect to the encapsulated data received.

Abstract: A method for distributed computing between a host computer and at least one accelerator device interconnected through a network includes profiling a data transfer rate and a computation rate for a range of data sizes to find an optimal chunk size for the data transfer through the network; splitting or aggregating a size of the data stored in a memory in the host computer for encapsulating the data into a chunk with the optimal chunk size; dispatching the encapsulated data to the accelerator device; and instructing pipeline computation to the accelerator device with respect to the encapsulated data received.

Abstract: A method for distributed computing between a host computer and at least one accelerator device interconnected through a network includes profiling a data transfer rate and a computation rate for a range of data sizes to find an optimal chunk size for the data transfer through the network; splitting or aggregating a size of the data stored in a memory in the host computer for encapsulating the data into a chunk with the optimal chunk size; dispatching the encapsulated data to the accelerator device; and instructing pipeline computation to the accelerator device with respect to the encapsulated data received.

Abstract: A method for executing kernels in a hybrid system includes running a program on a host computer and identifying in an instruction stream of the program a first instruction including a function of a target classification. The method includes generating a first kernel including the function and transmitting the first kernel to a client system to execute the first kernel based on identifying the first instruction as being of the target classification. The method also includes determining whether to store results of executing the first kernel in a read-only buffer of the client system based on determining whether a subsequent instruction of the target classification relies upon results of the first instruction.

Abstract: A method for executing kernels in a hybrid system includes running a program on a host computer and identifying in an instruction stream of the program a first instruction including a function of a target classification. The method includes generating a first kernel including the function and transmitting the first kernel to a client system to execute the first kernel based on identifying the first instruction as being of the target classification. The method also includes determining whether to store results of executing the first kernel in a read-only buffer of the client system based on determining whether a subsequent instruction of the target classification relies upon results of the first instruction.