End2End Performance - 2023.2 English

Result Prerequisite:

• Dataset: TLC Trip Record Data 2016-01 yellow taxi trip record. This dataset has 10,906,858 trip records, and column pickup_longtitude and pickup_latitude are used as spatial object inputs.
• Xilinx Device: U.2.
• Geopandas is run on server with Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz and PyGeos package applied.
• E2E Execution time includes read csv file to memory, compute distance, and sort distance.
• Top K = 5.
• Number of CSV Parser PU = 2 or 4: considering the fact that processing ability of single CSV Parser PU is maximum one byte per cycle, which is the bottleneck of design, 2 or 4 CSV Parser PUs are instantiated to enhance the performance.

	E2E Execution Time(s)
KNN(2 CSV Parser)	4.301
KNN(4 CSV Parser)	2.448
Geopandas(1 thread)	36.097