MVSym: Efficient symbiotic exploitation of HLS-kernel multi-versioning for collaborative CPU-FPGA cloud systems

Cloud Warehouses have been exploiting CPU-FPGA collaborative environments, where clients share the same infrastructure to maximize resource utilization with energy efficiency. In this scope, resource provisioning is challenging as kernels may be dispatched to CPU and FPGA in a scenario with highly variant resource availability and workload characteristics. In this work, we propose MVSym, a framework that leverages automatic HLS generation to enable further gains in CPU-FPGA Cloud. MVSym exploits the automatic generation from HLS to build libraries containing multiple versions of each incoming kernel request, greatly enlarging the available design space exploration passive of optimization by Cloud allocation strategies. MVSym makes kernel multi-versioning and allocation strategies work symbiotically, allowing fine-tuning regarding resource usage, performance, energy, or any combination of these parameters. Our framework achieves up to 4.62× makespan and 19.04× energy improvements over traditional strategies executing non-optimized kernels.