A smart grid model for high performance computing service

Many domains like deep learning, machine learning, and computational finance uses GPU's for decreasing the execution time. GPUs are widely used in data centers for high performance computing where virtualization techniques are intended for optimizing the resource utilization (e.g. GPU cloud computing). However, these data centers have some downsides like energy consumption and acquisition cost. We propose to take this resource virtualization a step forward by making use of user's computers that are connected to the internet that have powerful processing capabilities. The approach is similar to GPU cloud computing but our solution involves geographical distributed computer with heterogeneous hardware. Our model can be used as a service over the internet. The nodes in the service are represented by the number of active and available computers. The service will offer two types of load balancing, manual and automatic, letting the programmer to choose between them. In this way, we want to evaluate the feasibility of introducing a service that will provide processing capabilities through renting processing resources from domestic users.