search

kezhou2 / a-dda

Automatically exported from code.google.com/p/a-dda
0 stars 0 forks source link

Inaccuracies of timing of iterative solvers #107

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
There are two minor issues with timing of iterative solvers.
1) Communication time for initialization of the iterative solver is largely 
underestimated, since it does not include the communication time for 
matrix-vector product (for choosing the initial guess).
2) When BiCGstab stops in the middle of an iteration (intermediate convergence 
test succeeds), the total time of the iteration is not calculated and is thus 
effectively taken from the previous iteration. However, the calculation of the 
communication time for a single iteration is unaffected (used from the last 
iteration). In some cases, this inconsistency may lead to overestimation of the 
parallel efficiency (communication time about twice less).

Original issue reported on code.google.com by yurkin on 25 Aug 2010 at 3:09

GoogleCodeExporter commented 9 years ago 
This issue was closed by revision r956.

Original comment by yurkin on 25 Aug 2010 at 3:13