Gaussian-fall2012 - Center for High Performance Computing

Report
Using Gaussian & GaussView on
CHPC Resources
Anita M. Orendt
Center for High Performance Computing
[email protected]
Fall 2012
Purpose of Presentation
• To discuss usage of both Gaussian and
GaussView on CHPC systems
• To provide hints on making efficient use of
Gaussian on CHPC resources
• To demonstrate functionality of GaussView
http://www.chpc.utah.edu
Sanddunearch
156 nodes/624 cores
Infiniband and GigE
Updraft
256 nodes/2048 cores
Infiniband and GigE
85 nodes general usage
Owner nodes
Ember
382 nodes/45684 cores
Infiniband and GigE
67 nodes general usage
Turret Arch
Switch
Telluride
12 GPU nodes (6 CHPC)
Meteorology
Administrative
Nodes
IBRIX
NFS
NFS
Home
Directories
scratch systems
serial – all clusters
general – updraft
http://www.chpc.utah.edu
/scratch/ibrix/chpc_gen
ember, updraft
Getting Started at CHPC
• Account application – now an online process
– https://www.chpc.utah.edu/apps/profile/account_request.php
• Username unid with passwords administrated by campus
• Interactive nodes
– two per each cluster (cluster.chpc.utah.edu) with round-robin access to
divide load
• CHPC environment scripts
– www.chpc.utah.edu/docs/manuals/getting_started/code/chpc.tcshrc
– www.chpc.utah.edu/docs/manuals/getting_started/code/chpc.bashrc
• Getting started guide
– www.chpc.utah.edu/docs/manuals/getting_started
• Problem reporting system
– http://jira.chpc.utah.edu or email to [email protected]
http://www.chpc.utah.edu
Security Policies (1)
• No clear text passwords - use ssh and scp
• Do not share your account under any circumstances
• Don’t leave your terminal unattended while logged into
your account
• Do not introduce classified or sensitive work onto
CHPC systems
• Use a good password and protect it – see
gate.acs.utah.edu for tips on good passwords
http://www.chpc.utah.edu
Security Policies (2)
• Do not try to break passwords, tamper with files, look
into anyone else’s directory, etc. – your privileges do
not extend beyond your own directory
• Do not distribute or copy privileged data or software
• Report suspicions to CHPC ([email protected])
• Please see
http://www.chpc.utah.edu/docs/policies/security.html
for more details
http://www.chpc.utah.edu
.tcshrc/.bashrc
• Gaussian users need .tcshrc even if normally use bash – both are put in new
accounts
– Gaussian setup under individual compute cluster sections
– Uncomment out (remove # from start of line) EITHER the line that does a
source of g03.login OR the line for g09.login – they are mutually
exclusive!
• The script can be modified for individual needs with source .aliases at end
and the creation of an .aliases file
Gaussian Users Group
• User also needs to be in the gaussian group (check box on account
application form) – otherwise you will not have permission to run
gaussian/gaussview
– Command groups will show your groups; look for g05
http://www.chpc.utah.edu
Batch System
• All jobs run on compute nodes accessed through batch system with
Moab & Torque (PBS) as scheduler/resource manager
• More or less a first in – first out, but with backfill for best utilization of
resources
• Sanddunearch - no longer allocated - 72 hours max walltime
• Ember – 72 hours max walltime on CHPC nodes; long QOS available
– can also run in smithp-guest mode (#PBS –A smithp-guest); 24 hours max walltime and
preemptable on smithp nodes
• Updraft – 24 hours max walltime on CHPC nodes
– with allocation, can also run as preemptable (qos=preemptable in your #PBS –l line);
charged .25 the normal charges, but is preemptable
• No allocation – can still run in freecycle mode; this mode is preemptable on
updraft and ember. Automatic with no allocation - cannot choose
• Special needs/time crunch – talk to us; we do set up reservations
http://www.chpc.utah.edu
Job Control Commands
•
•
•
•
•
•
•
•
•
•
qsub script – to submit job
qdel job number – to delete job from queue (both waiting and running jobs)
showq – to see jobs in queue (add –r, -i, or –b, for running, idle, or blocked only)
– Use with | grep username to focus on your jobs only
– Idle jobs with reservations have * after job number
– If in blocked section there may be problems
qstat –a – PBS version of showq; has some different information (also –f jobnumber)
mshow –a – –flags=FUTURE – what resources are currently available to you
qstat –f job number – valuable info on deferred jobs
showstart job number – estimation of when your job will start (based on jobs ahead
of yours lasting for entire time requested); only works for jobs with reservations
checkjob (–v) job number – more detailed information; error messages at end
diagnose –n – shows you activity on nodes
More info on web pages and user guides
– http://www.chpc.utah.edu/docs/manuals/software/maui.html
– http://www.chpc.utah.edu/docs/manuals/software/pbs.html
http://www.chpc.utah.edu
Gaussian03
• Version E.01 (last version) installed
– /uufs/chpc.utah.edu/sys/pkg/gaussian03/E.01 for AMD (sanddunearch)
– /uufs/chpc.utah.edu/sys/pkg/gaussian03/E.01-EMT64 for Intel
(updraft/ember/PI-owned nodes on sanddunearch)
•
•
•
•
Main web site: www.gaussian.com
Have site license for both unix and windows versions
With G03, GaussView4 is standard
General information on CHPC installation
• http://www.chpc.utah.edu/docs/manuals/software/g03.html
• http://www.chpc.utah.edu/docs/manuals/software/gv.html
• Has information on licensing restrictions, example batch scripts, where to
get more information on the specific package
• Gaussian use is restricted to academic research only
http://www.chpc.utah.edu
Gaussian09
• Version C.01 current (still have B.01 and A.02 if needed)
– /uufs/chpc.utah.edu/sys/pkg/gaussian09/EM64T for Intel procs
/uufs/chpc.utah.edu/sys/pkg/gaussian09/AMD64 for AMD procs
• Have site license for unix version only
• Standard version of GaussView with G09 is GV5
• Chemistry groups purchased Windows license for both G09
and GV5
– Groups can purchase share to gain access
• General information on CHPC installation
• http://www.chpc.utah.edu/docs/manuals/software/g09.html
http://www.chpc.utah.edu
GaussView
• Molecular builder and viewer for Gaussian input/output files
• CHPC has campus licenses for linux version
– For Gaussian03 – standard is version 4
– For Gaussian09 – standard is version 5
• Access with gv & – provided you have uncommented the
Gaussian setup from the standard .tcshrc
• DO NOT submit jobs from within GaussView – instead create
and save input file and use batch system
• Examples of how to use to show MO’s, electrostatic potentials,
NMR tensors, vibrations given on Gaussian’s web page
http://faculty.ycp.edu/~jforesma/educ/
http://www.chpc.utah.edu
Highlights of G03/G09 Differences
• G09 does not use nprocl
– Limit of about 8 nodes due to line length issue
• New Restart keyword
– For property and higher level calculation restarts
– Still use old way for opt, scf restarts
• New easier way for partial opts/scans
– See G09 opt keyword for details
• New capabilities, new methods
– “What’s New in G09” at http://www.gaussian.com/g_prod/g09new.htm
• Improved timings
http://www.chpc.utah.edu
G03/G09 Script
• Two sections for changes:
#PBS -S /bin/csh
#PBS –A account
#PBS –l walltime=02:00:00,nodes=2:ppn=12
#PBS -N g03job
• And:
setenv WORKDIR $HOME/g09/project
setenv FILENAME input
setenv SCRFLAG LOCAL
setenv NODES 2
http://www.chpc.utah.edu
Scratch Choices
• LOCAL (/scratch/local)
–
–
–
–
–
Hard drive local to compute node
60GB on SDA; 200GB on updraft; 400GB on ember
Fastest option - recommended IF this is enough space for your job
Do not have access to scratch files during run but log/chk files written to $WORKDIR
Automatically scrubbed at end of job
• SERIAL (/scratch/serial)
– NFS mounted on all clusters (interactive and compute)
– 15 TB
• GENERAL (/scratch/general)
– NFS mounted on UPDRAFT compute nodes and all interactive nodes
– 3.5 TB
• IBRIX (/scratch/ibrix/chpc_gen)
– Parallel scratch file system (HP IBRIX solution)
– On UPDRAFT and EMBER compute nodes and on all interactive nodes
– 55 TB
http://www.chpc.utah.edu
Functionality
• Energies
– MM : AMBER (old one), Dreiding, UFF force fields
– Semi-empirical: CNDO, INDO, MINDO/3, MNDO, AM1,
PM3, PM6
– HF: closed shell, restricted and unrestricted open shell
– DFT: Many functionals, both pure and hybrid, from which
to choose
– MP: 2nd-5th order; direct and semi-direct methods
– Other high level methods such as CI, CC, MCSCF,
CASSCF
– High accuracy methods such as G1, G2 etc and CBS
http://www.chpc.utah.edu
Functionality (2)
• Gradients/Geometry optimizations
• Frequencies
• Other properties
– Population analyses
– Natural Bond Order analysis (NBO5 with G03)
– Electrostatic potentials
– NMR shielding tensors
– J coupling tensors
http://www.chpc.utah.edu
Input File Structure
• Filename.com
• Free format, case insensitive
• Spaces, commas, tabs, forward slash as delimiters between
keywords
• ! Comment line
• Divided into sections (in order)
– Link 0 commands (%)
– Route section – what you want calculation to do
– Title
– Molecular specification
– Optional additional sections
http://www.chpc.utah.edu
Input File: Link 0 Commands
• First “Link 0” options
– %chk
– %mem
– %nprocs
• Examples
– %chk=Filename.chk
– %mem=2gb
– %nprocs=4
• Note nprocl no longer used in G09
http://www.chpc.utah.edu
Number of Processors
• %nprocs – number of processors on one node
– sanddunearch – 4; updraft – 8; ember – 12
– There are owner nodes on sanddunearch with 8 processors
per node
http://www.chpc.utah.edu
Memory Specification
• Memory usage: default is 6MW or 48MB – all nodes
have much more than this!
• If you need more use %mem directive
– Units : words (default), KB, KW, MB, MW, GB, GW
– Number must be an integer
• Methods to estimate memory needs for select
applications given in Chapter 4 of User’s Guide
• %mem value must be less than memory of node
– Sanddunearch nodes have 8GB
– Updraft nodes have 16GB
– Ember nodes have 24GB
http://www.chpc.utah.edu
Input - Route Specification
•
•
Keyword line(s) – specify calculation type and other job options
Start with # symbol
– for control of the print level in the output file use #n, #t, #p for normal, terse, or more
complete output
– #p suggested as it monitors job progress; useful for troubleshooting problems
•
•
•
Can be multiple lines
Terminate with a blank line
Format
– keyword=option
– keyword(option)
– keyword(option1,option2,..)
– keyword=(option1,option2,…)
• User’s Guide provides list of keywords, options, and basis set notation
http://www.gaussian.com/g_tech/g_ur/l_keywords09.htm
http://www.chpc.utah.edu
Input - Title Specification
• Brief description of calculation - for user’s
benefit
• Terminate with a blank line
http://www.chpc.utah.edu
Input – Molecular Specification
• 1st line charge, multiplicity
• Element labels and location
– Cartesian
• label x y z
– Z-matrix
• label atom1 bondlength atom2 angle atom3 dihedral
• If parameters used instead of numerical values then
variables section follows
• Default units are angstroms and degrees
• Again end in blank line
http://www.chpc.utah.edu
Parallel Nature of Gaussian
• All runs make use of all core per node with nprocs
• Only some portions of Gaussian run parallel on multiple nodes
(includes most of the compute intensive parts involved with single
point energies/optimizations for HF/DFT)
• If time consuming links are not – job WILL NOT benefit from running
on more than one node
• Nmr and mp2 frequency are examples that do not run parallel; opt
and single point energies tend to scale nicely
• Not all job types are restartable, but more are restartable in G09 than
G03 (e.g., frequencies and NMR) – see new restart keyword
– Requires rwf from previous run
– Still restart optimizations and single point energies the old way
• CHPC does allow for jobs over standard walltime limit on ember if
needed – but first explore using more nodes or restart options
http://www.chpc.utah.edu
Timings G09 varying scratch system
All jobs on 1 ember node; run at the same time
on valinomycin (C54H90N6O18)
Scratch
setting
Medmp2
882 bf
536GB rwf file
Large opt
1284 bf; 51 opt steps
1GB rwf file
Large freq
of same
8GB rwf file
local
4.5 hrs
(used maxdisk)
10.25 hrs
10.5 hrs
serial
>30 hrs (ran
out of time)
27.5 hrs
14.75 hrs
ibrix
5.75 hrs
11.75 hrs
10.75 hrs
***depends strongly on amount of I/O and on other jobs usage on scratch system
http://www.chpc.utah.edu
Scaling G03/G09
B3PW91;650 bf; 8 opt steps; time in hours
# quad
nodes
wall time
# 8way
nodes
wall time
# 12way
nodes
wall time
1
23.25/18.5
1
5.75/5.5
1
3.25
2
12/9.25
2
3/3
2
1.75
4
6.25/5
4
2/1.75
8
4/3
8
1/1
16
3
http://www.chpc.utah.edu
DFT Frequency of same case
# quad
nodes
wall time
# 8way
nodes
wall time
# 12way
nodes
wall time
1
20.25/13.5
1
5.5/5
1
3.25
2
11/7.25
2
3/2.75
2
1.5
4
6/7.5
4
1.75/1.5
8
4.25/6
8
1/1
16
3.5
http://www.chpc.utah.edu
SDA/UP Scaling MP2 G03/G09
8 opt steps 338bf; time in hours (freq added)
# quad
nodes
wall time
# 8way
nodes
1
13.5
9.75(32.5)
10
5.5(27.5)
8.5
3(24)
1
2
4
wall time
# 12way
nodes
wall time
1
1(9.5)
1(7.5)
4(19)
2
2
1.5(16.5)
4
http://www.chpc.utah.edu
<1(7)
RWF Sizes – Choice of Scratch
• For a DFT optimization with 462 basis functions
– 150mb RWF
• For a DFT freq of above structure
– 1.1gb RWF
• For a MP2 optimization with 462 bf
– 55gb RWF AND 55gb SCR file
• For a MP2 frequency of above structure
– 247gb RWF
http://www.chpc.utah.edu
GaussView Demos
•
•
•
•
Building molecules
Inputting structures
Setting up jobs
Looking at different types of output
– Optimization runs
– Frequencies
– MOs
– Electron density/electrostatic potential
http://www.chpc.utah.edu
• Any questions – contact me
– [email protected]
– Phone: 801-231-2762
– Office: 422 INSCC
http://www.chpc.utah.edu

similar documents