Survey and Future Trends for FPGA Cloud Architectures
Hafsah Shahzad, Ahmed Sanaullah, Martin Herbordt
Abstract
In the last five years, FPGA presence in the cloud has gone from near zero (except for deeply embedded devices) to a large fraction of all high-end FPGAs sold. This is because FPGAs offer uniquely the performance, power, and flexibility needed to support the diversity and dynamicity of cloud workloads. We begin by observing that, although FPGAs are widespread, they cannot be randomly deployed as part of cloud infrastructure. Any FPGA cloud architecture must satisfy a number of constraints placed by the cloud provider. As a result, FPGA use in the cloud is non-uniformly distributed and motivated by the specific advantages and limitations that each unique architecture offers. In this survey, we provide an exploration and analysis of the trends in existing cloud FPGA architectures that highlight this complex relationship between architectures and system requirements. This allows us to identify novel architectures that are likely to offer substantial benefits for cloud workloads.