asychronous.paper

This repo holds the experimental design and results for Asynchronous execution paper

DeepDriveMD:

Asynchronous Execution setup:

Pipeline	Stage	NumTasks	Time	CPU	GPU
0000	Simulation	192	340.00091663355704	4	1
0002	Preprocessing	16	84.99926581009214	32	0
0002	MachineLearning	1	62.49933909684772	4	1
0002	Agent	96	37.49773836701004	16	1
0001	Simulation	96	340.01132125566215	4	1
0001	Preprocessing	16	85.00231602791855	32	0
0001	MachineLearning	1	62.49349059574567	4	1
0001	Agent	96	37.50304555141987	16	1
0003	Preprocessing	16	85.00230654482294	32	0
0003	MachineLearning	1	62.49658807469051	4	1
0003	Agent	96	37.50154022536458	16	1

Synchronous Execution setup:

Pipeline	Stage	NumTasks	Time	CPU	GPU
0000	Simulation	288	340.0031078322008	4	1
0000	Preprocessing	48	85.00433881721868	32	0
0000	MachineLearning	3	62.506966829048935	4	1
0000	Agent	288	37.503148763817244	16	1

DG1: Asynchronous Execution setup: Total: 768 Cores and 96 GPU (6 GPU per node)

Pipeline	Stage	NumTasks	Time	CPU	GPU
0000	T0	96	766.7707836629639	16	1
0000	T1T2	32	234.10703256865466	40	0
0001	T3T6	16	124.21534041883052	4	0
0002	T4T5	16	152.1652375323442	4	0
0002	T7	96	721.002649597601	4	1

Synchronous Execution setup:

Pipeline	Stage	NumTasks	Time	CPU	GPU
0000	T0	96	766.3557904708296	16	1
0000	T1T2	32	234.88397866417537	40	0
0000	T3T6	32	124.8144343585843	4	0
0000	T7	96	721.6538381216709	4	1

DG2: Asynchronous Execution setup:

Pipeline	Stage	NumTasks	Time	CPU	GPU
0000	T0	96	385.79899253812766	16	1
0000	T1T2	32	154.54474259009064	40	0
0001	T3T6	96	769.222570596073	4	1
0002	T4T5	16	230.56573660341817	4	0
0002	T7	16	460.71050717103435	4	0

Synchronous Execution setup:

Pipeline	Stage	NumTasks	Time	CPU	GPU
0000	T0	96	384.2774262654445	16	1
0000	T1T2	32	153.6810256298802	40	0
0000	T3T6	112	768.4886881624598	4	1
0000	T7	16	461.2054322747667	4	0