[ODK] Meeting 2018 10 24

Minutes for the October 24, 2018 meeting

1. Plots

• Effective cores vs speedup curves: this is nice, and fair. This is not linear, but this makes sense.
• For modular balanced with solve in, only 10 to 40 bits, no runs on 100 bits yet.
• No 500 bits on 32 cores, because run time was too long, which lead the server to force kill the process.

2. Exhaustive list of which runs to perform

Plots needed (as curves and histograms):

• Speedup against sequential runtime
• Real time

Perform runs on Dahu cluster, always on max cores (32) per nodes. All runs will use solve in on Modular Balanced. Matrices picked have run times roughly equivalent and are fairly challenging to process.

Matrix order	Bitsize	Nb Nodes	Done yet?
1000	1000	1	N
1000	1000	32	Y
1000	1000	64	Y
1000	1000	128	Y
1000	1000	256	N
2000	200	1	N
2000	200	32	Y
2000	200	64	Y
2000	200	128	Y
2000	200	256	N
4000	20	1	Y
4000	20	32	Y
4000	20	64	Y
4000	20	128	Y
4000	20	256	N
500	10000	1	N
500	10000	32	N
500	10000	64	N
500	10000	128	N

3. tl;dr

• Plot everything from section 2
• Good luck Clément!