Skip to main content

Table 3 Baseline assumption and notations used in the baseline study

From: Distributed large-scale graph processing on FPGAs

Symbol

Meaning

Con-stant value(if any)

\(\left|V\right|\)

Number of vertices

 

\(\left|E\right|\)

Number of edges

 

Sv

Size of a vertex (Byte)

4

Se

Size of an edge (Byte)

8

datawidth

Depending on the scenario (1024, 2048, 4096)

2048

\(P^{2}\)

The graph will be portioned into P\(\times\)P edge-blocks

 

Sblock

Size of an edge-block (MB)

10

fFPGA

Clock Frequency MHz

200

Tcomp

Time to process (read/Write to/from-chip BRAM)

 

Tcomm

Time to move data from host memory to FPGA

 

Tinit

Time to load the libraries (reduced after the first run)

 

TPCIe

Time to interconnect between boards

 

Tmem

Time to transfer data (loading blocks) between off-chip memory and PEs

 

Tbaseline

Time for CPU-only

 

TFPGA

Execution time in FPGA only

 

Sbram

BRAM storage size [it should be higher or equal to the size of each edge block Sbram ≥ Sblock(i,j)] (MB)

54

BWmem

DDR4-2400 (GB/s)

19.2

BWPCIe

PCIe 3.0 bandwidth ×16 (GB/s)

32

effPCIe

PCIe efficiency

0.8

effmem

Memory efficiency

0.8

SL

Number of SLRs in the FPGA die

4