M578 - Alexiades
Lab 7
Parallelizing your serial 2D axisymmetric code from Lab6
Following the same steps for parallelizing (and the experience of parallelizing your 1D code),
now parallelize your 2D code along the z direction.
Your parallelized code should produce identical results with your serial code (on MMr=MMz=32 mesh, factor=0.9).
With MMr = 64, try MMz = 256, 512, ... on nWRs = 2 , 4 , 8 , ...
If on a cluster, you should be seeing speedup!
In any case, calculate the Speedup and the Efficiency for each nWRs value ( S = T1 / Tp , E = S/p ).
Push it far enough (more nWRs) till the gain starts leveling off.
To stay on one node of ACF (16 cores, so up to nWRs=15), can take
Mz=1800 and run on nWRs = 2, 5, 10, 15 to calculate Speedup.
Submit the following:
a. max error (at tend=1) from a run with MMr=32, MMz=256 on nWRs= 4 workers.
--------------------------------------------------
b. The speedup and efficiency values you obtained in your runs.
A table of timings would be nice (on a fixed mesh, specify it):
nWRs timing Speedup Efficiency
Specify on what sort of system you made your runs.
--------------------------------------------------
c. lab7.zip with ONLY the essential files of your code
(z.* , Makefile , dat, how to run it)
!!! GOOD LUCK !!!
Hope you enjoyed the experience...
Now you can brag about it to your friends and in your CV !