CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums

Blog Entries

Register Blogs Members List Search Today's Posts Mark Forums Read

Old

Dark side of Amdahl's law. Epilogue

Posted January 12, 2012 at 11:03 by SergeAS
Updated January 13, 2012 at 08:32 by SergeAS

Dependence of efficiency on the number of cores is a very important characteristic of scalability solver based on which one can predict the scaling limit. For simplicity we consider only one option for the solver, using the asynchronous version of the MPI exchanges Isend() + Irecv(). It is seen that the dependence of the efficiency of parallelization of the solver is linear or nearly linear, with a gradual drop in effectivity with increasing number of cores used
...
Attached Thumbnails
Click image for larger version

Name:	HyperFLOW2D-effectivity0.png
Views:	818
Size:	10.4 KB
ID:	46   Click image for larger version

Name:	HyperFLOW2D-speedup0.png
Views:	313
Size:	10.3 KB
ID:	47  
SergeAS's Avatar
Member
Posted in Uncategorized
Views 985 Comments 0 SergeAS is offline Edit Tags
Old

Dark side of Amdahl's law. Part III

Posted December 9, 2011 at 12:04 by SergeAS
Tags mpi, parallel

...use a blocking call Sendrecv() instead of a pair of Send() + Recv() slightly improves the performance of the code but does not result to improved scalability


...
Attached Thumbnails
Click image for larger version

Name:	HyperFLOW2D-parallel-factor2.png
Views:	883
Size:	12.0 KB
ID:	42   Click image for larger version

Name:	HyperFLOW2D-speedup2.png
Views:	845
Size:	11.4 KB
ID:	43   Click image for larger version

Name:	HyperFLOW2D-speed2.png
Views:	851
Size:	11.8 KB
ID:	44   Click image for larger version

Name:	HyperFLOW2D-effectivity2.png
Views:	251
Size:	12.6 KB
ID:	45  
SergeAS's Avatar
Member
Posted in Uncategorized
Views 987 Comments 0 SergeAS is offline Edit Tags
Old

Dark side of Amdahl's law. Part II

Posted December 4, 2011 at 10:57 by SergeAS
Updated December 4, 2011 at 11:44 by SergeAS
Tags mpi, parallel

Let us consider in more detail how works domain decomposition in the simplest 1D case.
The computational domain is divided into several subdomains along one (usually the longest) dimension. The number of such subdomains usually corresponds to the number cores involved. The main data exchanges take place between neighboring subdomains. The size of transmitted data between adjacent subdomains in the general case depends on many factors (number of equations in the system, the size of the computational...
Attached Thumbnails
Click image for larger version

Name:	HyperFLOW2D-parallel-factor.png
Views:	257
Size:	11.3 KB
ID:	38   Click image for larger version

Name:	HyperFLOW2D-speedup.png
Views:	272
Size:	11.0 KB
ID:	39   Click image for larger version

Name:	HyperFLOW2D-speed.png
Views:	259
Size:	11.1 KB
ID:	40   Click image for larger version

Name:	domain.png
Views:	264
Size:	10.2 KB
ID:	41  
SergeAS's Avatar
Member
Posted in Uncategorized
Views 795 Comments 0 SergeAS is offline Edit Tags
Old

Dark side of Amdahl's law

Posted November 25, 2011 at 16:19 by SergeAS
Updated November 27, 2011 at 06:44 by SergeAS
Tags mpi, parallel

Looking the latest HPCSource I came across very well-known picture, illustrating the Amdahl's law:
S_p = \cfrac{1}{\alpha + \cfrac{1 - \alpha}{p}}


which shows the dependence of the speedup parallel code on number of processor cores for different fraction of time (\alpha \ne 0) spent in the part that was not parallelized. This diagram is interested in me for the...
Attached Thumbnails
Click image for larger version

Name:	HyperFLOW2D-speedup.png
Views:	281
Size:	7.7 KB
ID:	31   Click image for larger version

Name:	HyperFLOW2D-speed.png
Views:	279
Size:	7.2 KB
ID:	32   Click image for larger version

Name:	HyperFLOW2D-parallel-factor.png
Views:	276
Size:	6.5 KB
ID:	33   Click image for larger version

Name:	VonKarmanVorticeStreet2.gif
Views:	275
Size:	87.4 KB
ID:	36  
SergeAS's Avatar
Member
Posted in Uncategorized
Views 978 Comments 2 SergeAS is offline Edit Tags

All times are GMT -4. The time now is 05:51.