Formula | Description |
---|---|
\(\text {P}=\left| \sqrt{\frac{S_{e}\times \left| E \right| }{S_{block}}} \right|\) | Total number of partitions based on GridGraph preprocessing |
\(Size_{data} = S_{block} \times {P^2}\) | Edge block size to be processed |
\(BW_{BRAM} = \frac{data_{width} \times {f_{FPGA}}}{S_{e}}\) | BRAM Bandwidth calculation |
\(T_{mem} =\frac{\left| V \right| \cdot S_{v} + \left| E \right| \cdot S_{e}}{BW_{mem} \times {eff_{mem}}}\) | Load time for Vertex and Edges data from/to Global Memory to/from PEFPGA |
\(T_{PCIe} = \frac{size_{data}}{BW_{PCIe}\times {eff_{PCIe}}}\) | PCIe transfer time |
\(T_{comm} = T_{init} + T_{PCIe} + T_{mem}\) | Transfer time from/to CPU host to/from PEFPGA |
\(T_{FPGA} = T_{comm} + T_{comp}\) | Total time for FPGA |
\(T_{comp} = \frac{\left| V \right| + \left| E \right| }{BW_{BRAM}}\) | Computation time in the FPGA |
\(S = \frac{S_{baseline}}{S_{FPGA}}\) | Speedup over the baseline execution time |