The Future
The future of high-performance and scientific computing is in the cloud, with accelerators including FPGAs providing reconfigurable hardware that can be adapted to a user's applications. The trend is to replace server-centric architectures with a disaggregated data center, where memory and accelerators are connect directly to the network. Disaggregation removes many existing bottlenecks and ensures better utilization of resources. It also opens up new research directions, including assigning and scheduling jobs to processors, be they CPUs, GPUs or FPGAs as well as writing applications that target each type of accelerator. Security is also a challenge when there are many devices directly connected to the network and to one another. Our research aims to make FPGAs in the cloud available to a large number of researchers. Cloud users of the future will be able to take advantage of a computer platform with fewer memory bottlenecks and the processing power to deliver graph processing, machine learning, security and privacy applications on large data as well as accelerating scientific applications and new applications not yet imagined. With OCT, systems researchers have a platform where they can provide FPGAs as a Service (FaaS) and experiment with what that should look like.