微波EDA网,见证研发工程师的成长!
首页 > 研发问答 > 微波和射频技术 > 天线设计和射频技术 > Using HFSS on Supercomputers using PBS

Using HFSS on Supercomputers using PBS

时间:04-04 整理:3721RD 点击:
Hi,

I am trying to submit an HPC job for HFSS 16.1 using a pbs job. However, it only seems to be using one node and is insanely slow. Can somebody please look into my pbs job below and let me know what I am doing wrong.

It needs 800GB RAM at a minimum

Code C - [expand]
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
#!/bin/bash -l
 
#PBS -N SRR_jcho
#PBS -j oe
 
#PBS -l walltime=24:00:00,nodes=4:ppn=16,mem=1000gb
#PBS -m abe
#PBS -M [email]agarw071@umn.edu[/email]
 
 
module load hfss
 
export OptFile=S{PBS_O_WORKDIR}/Options.txt
export ANSYSEM_JOB_ID=S{PBS_JOBID}
export ANSYSEM_HOST_FILE=SPBS_NODEFILE
export LINUX_SSH_BINARY_PATH=/usr/bin
export ANSYSEM_LINUX_HPC_UTILS=/opt/software/AnsysEM/15.0/AnsysEM15.0/Linux64/schedulers/utils
 
cd S{PBS_O_WORKDIR}
    # mkdir -p S{PBS_O_WORKDIR}/scratch
 
echo creating batch options list
echo \Sbegin \'Config\' > S{OptFile}
    echo \'HFSS/NumCoresPerDistributedTask\'=S{PBS_NUM_PPN} >> S{OptFile}
    echo \'HFSS/HPCLicenseType\'=\'Pool\' >> S{OptFile}
echo \'HFSS/SolveAdaptiveOnly\'=0 >> S{OptFile}
    echo \'HFSS/MPIVendor\'= \'Intel\' >> S{OptFile}
echo \'HFSS-IE/NumCoresPerDistributedTask\'=S{PBS_NUM_PPN}  >> S{OptFile}
echo \'HFSS-IE/HPCLicenseType\'=\'Pool\' >> S{OptFile}
    echo \'HFSS-IE/SolveAdaptiveOnly\'=0 >> S{OptFile}
echo \'HFSS-IE/MPIVendor\'=\'Intel\' >> S{OptFile}
    # echo \'tempdirectory\'=\'S{PBS_O_WORKDIR}/scratch\' >> S{OptFile}
    echo \Send \'Config\' >> S{OptFile}
chmod 777 S{OptFile}
 
hfss -Ng -monitor -distributed -batchoptions "HFSS/HPCLicenseType=pool" -BatchSolve vacuum_gap_10nm.hfss

Hi kritia,

I was trying something similar not long ago, and was getting the same results. It turned out that the meshing process took upwards of 24 hrs, and this process only used one node.

What I ended up doing is pre-meshing the simulation on a desktop computer and then transferring the simulation data over to the supercomputer to do the actual adaptive passes.

This was with a previous version of HFSS (12, I think), so as a disclaimer they may have fixed this and your problem could be entirely unrelated.

Good Luck!

Copyright © 2017-2020 微波EDA网 版权所有

网站地图

Top