Energy Characterization of Graph Workloads Data

The EnergyCharacterizationAverageData.csv file contains the GAP benchmark kernels' energy characterization data. The detailed data analysis can be found at:

• Ankur Limaye, Antonino Tumeo, and Tosiron Adegbija, "Energy Characterization of Graph Workloads," in Sustainable Computing: Informatics and Systems Journal.
• The paper was presented at the 11th International Green and Sustainable Computing Conference (IGSC '20), 19 -- 22 October 2020.

Experimental Setup

Our experimental setup was:

• Six GAP benchmark kernels -- bc, bfs, cc, pr, sssp, and tc
• Twelve input graphs --
  • 9 synthetic Kronecker graphs with varying scales and degrees (kron-g20-k4, kron-g20-k8, kron-g20-k16, kron-g22-k4, kron-g22-k8, kron-g22-k16, kron-g24-k4, kron-g24-k8, and kron-g24-k16)
  • 3 real-world graphs: road, twitter, and road
• Four system configurations -- Single-socket Multithreading (SS-MT), Single-socket Simultaneous Multithreading (SS-SMT), Dual-socket Multithreading (DS-MT), and Dual-socket Simultaneous Multithreading (DS-SMT)

Data Format

Each column of the CSV file contains the performance counter data for each GAP kernel executing various input graphs for the different system configurations. The key column contains the entries in the following format:

Kernel---Graph---#Threads---#Processors---Threads/Core---Simultaneous_Multithreading_Enabled?

For example, bc-kron-g20-k4-t1-1s-1tpc-nh implies bc kernel executing the kron-g20-k4 input graph, with #threads = 1, #processors = 1, and simultaneous multithreading disabled (i.e., 1 thead per core)