Running Time Anomalies in solution of HW2

[arijit]

Hi,

1) I was wondering if the other groups are also experiencing the
following type of running time anomalies. For a fixed value of k, as
we increase p, the running time also increases; i.e.


k=96, p=1 Elapsed time during CGSOLVE: 0.122765
k=96, p=2 Elapsed time during CGSOLVE: 0.159117
k=96, p=4 Elapsed time during CGSOLVE: 0.423924
k=96, p=8 Elapsed time during CGSOLVE: 0.603213
k=96, p=12 Elapsed time during CGSOLVE: 0.756158
k=96, p=16 Elapsed time during CGSOLVE: 0.934735
k=96, p=32 Elapsed time during CGSOLVE: 1.741897

Is it because the data transferring time dominates over the parallel
computation time?

2) Currently my available balance shows 346. I guess, this is the
utilization time left in minute; and it will not keep on decreasing if
someone does not use the triton a/c for some period of time.

Thanks,
Arijit

[Kyle Klein]

Try scaling on a larger problem size. The problem you are running on is so small that the startup communication cost is dominating the the actual computation that needs to be done (or so I think).

best,

Kyle

--
Kyle Klein 

Ph.D. Student 

Department of Computer Science UC Santa Barbara

http://www.cs.ucsb.edu/~kyleklein

[arijit]

Thanks Kyle...You are right....there is no running time anomaly for
higher values of k.

Arijit