Context Navigation

Posts for the month of July 2009

Parallel on the cluster

The table below shows sweep times for the reactive wetting simulation on the cluster (Simulation 122). The results are awful as was probably to be expected due to the bandwidth issues with the cluster.

Nproc	sweep 1	sweep 2	sweep 3
1	17.43	13.65	13.72
2	9.65	7.34	7.40
4	19.23	19.76	7.26
8	23.11	20.72	21.44
16	29.33	100.40	126.25

Posted: 2009-07-28 16:49 (Updated: 2009-07-28 16:50)
Author: wd15
Categories: (none)
Comments (0)

Simulation 121

Simulation 121 is a fine grid version of the base case. Copied from 119 with the grid density doubled.

Posted: 2009-07-28 14:56 (Updated: 2009-07-28 14:57)
Author: wd15
Categories: (none)
Comments (0)

Simulations on Raritan

We copied simulation 119 to 120 to remove the HDF5 calls. At this stage we don't have HDF5 installed due to build issues with szip

Commands for raritan

This submits a job with 4 nodes and 1 processor per node

qsub -q infiniband -lnodes=4:ppn=1 -W x=NACCESSPOLICY:SINGLEJOB ./myscript

Simulations

8-128 processors

8-32 processor per node basis

dt from 1e-15 to something

Problem size 256x256, 512x512, 1024x1024

Simulation set 1, $\Delta t = 1 \times 10^{-15}$ , 256x256

$N$ : number of nodes (each node has X processors) $P$ : number of processors being used on each node $t_i$ : wall clock time at sweep $i$

		$t_{1}$	$t_{2}$	$t_{3}$	$t_{4}$
4	1

Posted: 2009-07-21 10:27 (Updated: 2009-07-21 10:51)
Author: wd15
Categories: (none)
Comments (22)

Equilibrium calculations

Currently running 1D equilibrium calculations to determine thermodynamics which has an equilibrium contact angle of $\Pi / 18$ . This is done by varying $B$ . Simulations 116, 117 and 118 are being used for this calculation. B_0 = 25418.6582832