Acceleration prediction in hybrid systems

A computer readable storage device storing a program of instructions executable by a machine to perform a method of predicting acceleration in a hybrid system, the method comprising:

• receiving response time and CPU utilization measurements in first host-only computer architecture running a workload;
• performing parameter inferencing on the response time and CPU utilization measurements of the first host-only computer architecture running a workload;
• determining service time of a host machine in the first host-only computer architecture running the workload based on the parameter inferencing of the response time and CPU utilization measurements of the host-only computer architecture running a workload;
• receiving response time and CPU utilization measurements in first host-accelerator computer architecture running the workload;
• performing parameter inferencing on the response time and CPU utilization measurements of the first host-accelerator computer architecture running the workload;
• determining service time of a host machine in the first host-accelerator computer architecture running the workload based on the parameter inferencing of the response time and CPU utilization measurements of the first host-accelerator computer architecture running the workload;
• determining service time of an accelerator in the first host-accelerator computer architecture running the workload based on the parameter inferencing of the response time and CPU utilization measurements of the first host-accelerator computer architecture running the workload;
• determining aggregate communication delay between the host machine and the accelerator in the first host-accelerator computer architecture running the workload;
• determining a number of cross system calls in the first host-accelerator computer architecture running the workload;
• determining a communication delay associated with each of the cross system calls based on at least the aggregate communication delay between the host machine and the accelerator in the first host-accelerator computer architecture running the workload, and the number of cross system calls in the first host-accelerator computer architecture running the workload;
• determining host machine overhead and accelerator overhead in the first host-accelerator computer architecture associated with each of the cross system calls based at least on the service time of a host machine in the first host-only computer architecture running the workload, the service time of a host machine in the first host-accelerator computer architecture running the workload, the service time of an accelerator in the first host-accelerator computer architecture running the workload, and the number of cross system calls in the first host-accelerator computer architecture running the workload;
• and
• predicting an application response time for a candidate application to be run in a second host-accelerator computer architecture, based at least on the determined host machine overhead, the accelerator overhead, and the communication delay associated with each of the cross system calls.