UBC-ECE-Sasha / PIM-encryption

Block decryption in memory
2 stars 2 forks source link

Experiment environment details need to be included with data #9

Open jpzg opened 4 years ago

jpzg commented 4 years ago

A bunch of parameters - DMA transfer size, optimization level, specifics of the code - affect performance. But the outputted data doesn't have that context, and the context could get lost when the data is copied to other places. Everything that determines the results (tuning parameters, compilation options, code version, maybe even output from things like uname, /proc/cpuinfo, DPU diagnostics and SDK version, what else is running, etc.) should be included in a file with the results, or maybe in each file (if they're updated separately).

jpzg commented 4 years ago

The CONF variable in the makefile does something like this - it creates a file in build/ with the number of tasklets and DPUs in its name. However, that doesn't include all the necessary details and is outdated (the number of DPUs is not configured at compile-time anymore).

(edit: the CONF file also causes make to recognize a change and rebuild on a change of parameters, as well as a change in a file. That's probably the main intention, rather than storing compilation details.)