CoMerge

Experimental setup for exploration of efficient data placement in shared heterogeneous memory systems.

Hybrid Memory System Emulation

Scripts in throttle/

Overall Application Sensitivity

run_all_throttle.sh in the individual application subfolders, to see the overall application performance across different combinations of bandwidth and latency NVM emulation.

Data Structure Sensitivity

Needs custom modification of 1 line of code in each source code file. The line refers to the choice of memory node for the allocation of the data object, and is clearly marked.

Collocation Analysis

For specific data tierings and NVM emulation, implemented as stated in the two previous steps, run in parallel the desired applications, multiple times.

Experimental raw data

In folder data/

Reference

For further detail and experimental result analysis refer to:

Thaleia Dimitra Doudali and Ada Gavrilovska. 2017. CoMerge: Toward Efficient Data Placement in Shared Heterogeneous Memory Systems. In Proceedings of MEMSYS 2017, Alexandria, VA, USA, October 2–5, 2017, 11 pages.

Addendum

I am extending Thaleia's original work to enable testing of Intel Optane DC Persistent Memory

fsgeek/CoMerge-extended