Ocean Modeling Discussion

shchepet

...and, of course, all the disappointing experience from the past still applies as it was just yesterday night viewtopic.php?p=7771

shchepet

AMD Ryzen Threadripper-pro-5995wx may be a very attractive and promising option

https://www.amd.com/en/products/cpu/amd-ryzen-threadripper-pro-5995wx

but I have no personal experience with them to advise.

In any case, memory bandwidth is always a bottleneck, so the more memory channels you ...

shchepet

Generally speaking, with limited money you can get better performance vs. $$$ cost, if you with "enthusiast" class PC. Get a high-level gaming mother board. Quad-channel memory system in a MUST. Do not settle with dual-channel memory. Pick reasonably priced Core i7-E or i9 processor (may be AMD ...

shchepet

Is this an idealized problem?

If yes, you have to make a your own routine which creates these field analytically.

As an example (actually a fairly old one) you may use

http://people.atmos.ucla.edu/alex/ROMS/IslandWakeNotes.pdf

The technique is essentially the same as using side boundary ...

shchepet

....The goal is to have the flow fully developed over the alongshore direction....

You want periodicity and simultaneously want to maintain your alongshore velocity be close to a desired value. This is over-specification. You do need periodic boundaries to do that. Instead you need to specify you ...

shchepet

What is the acceptable level of orthogonality errors? I would say none. Or, more
precisely, the question is irrelevant, because orthogonality errors can be kept at
the level roundoff errors of double precision -- it is technologically achievable,
and therefore should be done this way. Then there is ...

shchepet

The answer in "no", all three of them, A,B,C, cannot be made straight at the same time.

This is the nature of conformal mapping: once A and C are straight, there is no way to fit a
curve which would go along the coast, and be perpendicular to A and C at the junction points.

B can me made ...

shchepet

Recently I was asked to help set up ROMS grid for a lake which has its bathymetry data available only as
a set of point measurements from a boat: the depth was measured by echo sounder accompanied by GPS,
so the data comes as a file of the following structure, three numbers on each line

latitude ...

shchepet

No, you should be able to compile the package using any compiler.

I updated the package (see attachment at the end of this post) specifically paying attention to
the warnings issued by GCC compiler, which was forced into checking conformance to Fortran 2008
standards in pedantic mode. No bugs ...

shchepet

Hi Steve,

Remember me in Florida State back from 1992-1993? It was a nice time back then.

You problem will take only a few minutes to overcome because the very same package
you have compiled contains command-line operators nc_add_att and nc_del_att which
allow you to put all the necessary ...

shchepet

Obviously this should be

C$OMP PARALLEL SHARED(Lm,Mm, pm,pn, f,f_q, fnd_rmask,rmask, dm_u,dn_v,
C$OMP& iA_q)
call init_vort_thread(Lm,Mm, pm,pn, f,f_q, fnd_rmask,rmask,
& dm_u,dn_v, iA_q)
C$OMP END PARALLEL

i.e., change "& into "C$OMP&" in continuation line for OpenMP directive
in the original ...

shchepet

Try the tool editmask which is in the myroms/landmask folder of Matlab tools

John,

Here is the problem: At first many thanks to Andrey Scherbina for creating editmask tool back in 1999.
I checked editmask in myroms/landmask today by getting it from via svn,
svn checkout --username shchepet ...

shchepet

Error: Symbol ‘nf_netcdf4’ at (1) has no IMPLICIT type

This one you should be able to overcome yourself. At first, my package comes with several
README.something files: just look through them. Among them is README.compiling which
tells what to do.

Then, it is quite obvious that your netcdf.inc ...

shchepet

you can fit weird shapes after some practice...

: Black Sea grid.

shchepet

The reason why you have these compiling problems is a faulty version of mpc.F supplied
with the package at some point in the past. This was corrected, but it appears that the
old version came back because at some point the web page was restored from the backup
after disk failure and I did not ...

shchepet

The procedure for generating ROMS land mask from USGS coastline data described here:
viewtopic.php?f=23&t=3878&p=14908

shchepet

Is this the kind of grid you want to generate?

shchepet

What is the motivation for doing this? I inspected WRF file and it looks like this
is just another layer of software toward black-boxing it one level further. What is
wrong with standard build procedures?

shchepet

Jamie,

This is very useful as you bringing it in, and actually encouraging. The first time
I heard about AMD Ryzen is from my son, who is a kind of enthusiast, but I have no first
hand experience.

Over long period of time Intel and AMD had very different design philosophies: let's
just assume that ...

shchepet

The profile above indicates that the most computationally expensive part is seawater Equation of State. Is this actually true?

shchepet

There are several of them associated with different processes.

1. Barotropic mode. The easiest to predict and control -- based on speed of
the external waves, c_g=sqrt(g*h) where h is local depth, then
Cg = c_g*dtfast*sqrt{1/dx^2 +1/dy^2} with dx and dy
evaluated locally; Cg_max=max{Cg} over all ...

shchepet

.....16 cores. 4*4 grid: wall time 17s
.... 16 cores. 4*4 grid: wall time 27s
.... 64 cores (KNL). 8*8 grid, wall time 28s

Am I interpret correctly that "4*4 grid" means tiling 4x4? If yes, read
https://www.myroms.org/forum/viewtopic.php?f=17&t=2001&p=7771
-- used to be hot topic sometime in ...

shchepet

....is there a website you could direct me to?

please read
http://people.atmos.ucla.edu/alex/techn ... g_sequence

shchepet

Another problem I noticed is

Operating system : Linux
CPU/hardware : x86_64
Compiler system : gfortran
Compiler command : /usr/bin/gfortran
Compiler flags : -frepack-arrays -fopenmp -O3 -ffast-math -ffree-form -ffree-line-length-none -ffree-form -ffree-line-length-none

which means that you use ...

shchepet

The errors you are reporting are unrelated to netCDF installation.

The error message from your roms model says it all:

Resolution, Grid 01: 0041x0080x016, Parallel Threads: 2, Tiling: 001x001

ROMS/TOMS: Wrong choice of domain 001 partition or number of parallel threads.
NtileI*NtileJ must be a ...

shchepet

This means that XML packages from Linux distribution are not installed.
Type (assuming that you have rpm-based Linux; Debian uses different command)
rpm -qa | grep XML
you should get something like this

perl-XML-SAX-0.990.0-7.mga5
perl-XML-SAX-Base-1.80.0-5.mga5
perl-XML-NamespaceSupport-1.110.0 ...

shchepet

Read
http://people.atmos.ucla.edu/alex/techn ... g_sequence

shchepet

I tried to build grid for lake Shira using the latest available GSHHS version 2.3.6
shoreline data set taken at full resolution, gshhs_f.b, see
https://www.soest.hawaii.edu/pwessel/gshhg/
and this is what I managed to get:
http://people.atmos.ucla.edu/alex/tmp/shira_lake1.png
The lake exists in ...

shchepet

Most (most likely all) problems of this kind associated with linking to netCDF
library can be avoided if you compile netCDF (including dependencies hdf5, zlib,
etc) libraries yourself using matching compiler and compiler flags (the same as
you use for ROMS) rather than "install" it from your Linux ...

shchepet

Kate, this actually exposes a long outstanding problem: how useful ROMS diagnostic of PE?
And what would be the most meaningful definition of PE in principle? Yes, the computation of
PE diagnostic was there in diag.F from the very beginning inherited from SCRUM and SPEM,
yet nobody was wiling to ...

shchepet

There is one more thing to add to this topic specifically related to the fact that you may
want not just geostrophic balance, but the geostrophic balance as it would be felt by the
discrete model , i.e., ROMS.

ROMS uses C-grid, which means that discrete u s and v s in Coriolis terms are
subject to ...

shchepet

I set velocity field first and calculate the temperature field through geostrophic balance

It is usually works the other way around: density field + free surface (via level of no motion or some other idea of this sort) ==> pressure field ==> pressure gradient ==> velocity field

shchepet

...In the process, I have rewritten the ROMS source code of the test case “UPWELLING”
Obviously, going from upwelling toy problem to a full-complexity, realistic regional nested
Kuroshio simulation is a long way to go, and you have to rewrite a lot. But if the question
is what pre-configured ...

shchepet

Speaking CF-conventions....

Would it be hard to adapt and universally enforce a convention that time dimension and
the associated timing variable in any ROMS-related netCDF file have the same name ?

This should apply to ROMS code itself as well as all pre- and post-processing software,
Matlab, and ...

shchepet

The rationale for computing angle is very simple:

Consider a small Dxi increment along the direction of ROMS grid curvilinear coordinate
xi: the associated displacement in longitudinal direction measured in km is
R * cos(Lat) * dLon/dxi * Dxi
where R is Earth radius; displacement in latitudinal ...

shchepet

See the bottom half of the following routine.

Angle can be either in degrees or radians -- it is just a matter of personal preference.
ROMS code itself does not use angle, but various pre- and post- processing routines use
it to rotate vectors -- velocities if one needs to prepare files for ...

shchepet

Now, I have the following file: roms_clm_HYCOM_Y2015M10.nc 34GB

Another thing to mention here is that 64BIT_OFFSET netCDF format has virtually no restriction
of the file size, however to exceed 2 GByte limit the file MUST HAVE UNLIMITED DIMENSION, and

(1) all variables not having unlimited ...

shchepet

Kate (and Angelolemos as well),

You should be more specific here: rather than advise to post it on AGRIF forum rather than
here, or trying to motivate AGRIF people to be more flexible (good luck with this, huh...)

The error
GET_TCLIMA - unable to open climatology NetCDF file: ../SCRATCH/roms ...

shchepet

Can the existing nesting mechanism in ROMS handle such corners?

No, because coordinates become strongly non-orthogonal there, so equation require
extra terms which ROMS lacks. ROMS generally follows Arakawa and Lamb, 1977 orthogonal
curvilinear coordinate framework and the associated ...

shchepet

I understand the feeling of excitement when one is about to embark on a new
adventure, but actually, while cubed-sphere was famously used for many years
in ECCO model (this is MITgcm global configuration), the most recent version,
ECCO v.4, has abandoned this approach in favor of conformal mapping ...

shchepet

Tomasz, you must be more specific: Hernan points to the differences in
step2d which dated back to long time ago as reflected in a post from 2005,
http://www.myroms.org/forum/viewtopic.php?f=19&t=280
still relevant today ( ? ) -- simply put, AGRIF code was updated with respect
to this matter, but ...

shchepet

Generally you want to keep only what is mathematically necessary in the file,
so not only u_mask, and v_mask should be discarded, but few other items as well.

For a spherical grid this is all what you need:

netcdf pac16_grid {
dimensions:
xi_rho = 1330 ;
xi_u = 1329 ;
eta_rho = 1218 ;
eta_v ...

shchepet

Check your numbers:

.....
x=(xr(i,j)-x0) ; y=(yr(i,j)-y0)
alfa=(x*x+y*y)/200.0_r8
zeta(i,j,1)=EXP(-alfa)
....

where 200.0_r8 means 200 square meters, or same as (14 m)^2, meaning that the semi-width of your Gaussian is only 14 meters. However your

Xsize=1000.0E+03_r8
Esize=500.0E+03_r8 ...

shchepet

An analytical setup for testing baroptropic Flather- or Orlanskii-type boundary
conditions from article Mason, et al., 2010: Procedures for offline grid nesting
in regional ocean models, Ocean Modeling, vol. 35, pp. 1-15, especially Sec. 2.1.2
and Figure 3 there (expanding ring in free surface field ...

shchepet

Dear All,

A new automatic tool to construct ROMS land mask from GSHHS
(Global Self-consistent Hierarchical High-resolution Shorelines) dataset
is now available as a compile-once -- use-forever command-line operator.
This is the preferred the way and, in fact superior to the more usual practice
of ...

shchepet

OK. Let's just summarize it. I just downloaded the latest version, ROMS 3.7 svn rev. 726, and the relevant piece of
code looks like

cff=1.0_r8/(0.5_r8*(GRID(ng)%h(Iend ,j)+ & ! <-- A
& GRID(ng)%h(Iend+1,j)))
cff1=SQRT(g*cff) ! <-- B
Cx=dt2d*cff1*0.5_r8*(GRID(ng)%pm(Iend ,j)+ & ! <-- C
& GRID ...

shchepet

The original code in u2dbc_im.F is as follows, and it is correct:

cff=0.5*(h(iend,j)+h(iend+1,j))
hx=sqrt(g/cff)
cx=dtfast*cff*hx*0.5*(pm(iend,j)+pm(iend+1,j))

zx=(0.5+cx)*zeta(iend,j,kstp)+(0.5-cx)*zeta(iend+1,j,kstp)
if (cx > 0.292893218813452) then
zx=zx + ( zeta(iend,j,knew) +cx*zeta ...

shchepet

The following message
forrtl: severe (408): fort: (2): Subscript #1 of the array GRIDNUMBER has value 1 which is greater than the upper bound of 0
says it all, you either have a simple code bug, or an uninitialized index (most likely also due to code bug, or, alternatively, due to allocation ...

shchepet

There are in principle two way to handle this situation, which should not be mixed together.

1. The first one is as you suggested: switch first_time should have exactly the same status
as any global summation variable, which mans that it should be shared, that is not declared
as threadprivate, and ...

shchepet

Pay attention to ncid = 0 in

NETCDF_CLOSE - error during closing of file, ncid = 0

most likely you are trying to close a file which was not opened
at the first place.

shchepet

libifport.so.5 is a part of Intel compiler installation and it belongs to
the standard library directory within ifort installation tree, say, go to
the directory where your intel compuler(s) are installed and type

find . -name libifport.so.5 -print

The coutcome should be something like ...

shchepet

If this would help, there is a fortran code designed to work as a command-line operator
to transfer topography from Gebco (as well as other datasets like SRTM30) to ROMS
grid file. http://www.atmos.ucla.edu/~alex/ROMS/tools.tar

shchepet

Your choice Vtransform=2 ; Vstretching=4 in combination with such small value for critical depth

TCLINE=5.0d0

would yield an very tight packing of sigma levels toward the surface: suppose you have 40 layers
total. Then your uppermost grid box will be only 12.5cm thick, and so do other grid boxes ...

shchepet

The cause of your problem is the line in your makefile

LIBS := -L$(NCARG_ROOT)/lib -lncarg -lncarg_gks -lncarg_c

which is way outdated because this is how NCAR Graphics 4.4 used to be.

Basically the above line does not specify all the libraries NCAR graphics relies upon.

What version of NGAR ...

shchepet

You have to do both: look for artificial reflections, and look for maximum
allowed time step (actually to make sure that your boundary algorithm does
not impose additional restrictions on dtfast relative to what is the natural
limit of the time stepping algorithm), as well as check how the ...

shchepet

The problem with these values is that they are way all too small.

Normally, when you are running the code in 3D configuration you should not waste your CPU time on the barotropic mode, so the time step dt and mode splitting ratio should be set accordingly from the respective stability limits ...

shchepet

What is your barotropic gravity wave Courant number in these experiments?

Cg_max = ?

shchepet

Besides everything else highlighted by David, your code spends ~55% of time computing
barotropic mode. This is way too much. What is your barotropic gravity wave Courant number,
Cg_max=max_{i,j}{ dtfast*sqrt[g*h*(1/dx^2+1/dy^2)]} computed by "metrics.F" and reported
as "Maximum barotropic Courant ...

shchepet

There is no bug here as far as I can see.

cff1=gmax*Zoo*Phyt/(kp + Phyt*Phyt)
cff3=1/(1+cff1)
Phyt(new)=Phyt(old)*cff3
Zoo(new)=Zoo(old)+Phyt(new)*cff1

The whole biological time stepping procedure is dated back into 1998 -- about 14 years ago, see http://www.atmos.ucla.edu/~alex/biology , and ...

shchepet

Grid size 960*768*30 ... tiling 8*32 ... Cpu: four 64-core AMD Opteron ...mvapich2 1.6,
Let me guess: Based on what said above, you are running code in MPI mode using
4 hardware nodes each of whom has 4 CPU sockets, and each CPU is a 16-core Opteron.
The hardware nodes are interconnected by ...

shchepet

Because at this moment t(i,j,k,nnew,itrc) has the meaning of tracer concentration
multiplied by Hz(i,j,k) . Since Hz changes because of free surface evolution, tracers
are multiplied by Hz at time step "n" before the barotropic time stepping begins for
the current 3D time step, then all fluxes ...

shchepet

Watching this conversation between Mark and Hernan, I realized today that it has been approximately
10 years (no kidding!) since I eliminated all C$OMP PARALLEL DO directives from my code. Back then
I was motivated by making the code more explicit (and, frankly by beauty as well), but it appears ...

shchepet

Hernan,

by adding a small increment dt to time as follows from your code,

...
time(ng)=time(ng)+dt(ng)
...

(see close to the end) you are back to the original problem: you accumulate roundoff errors in counting time.

What you should do instead is to change the above into something of the ...

shchepet

This version of stretched terrain-following coordinate formally appears first as Eqs. (2.2)-(2.4) in

Shchepetkin, A. F. and J. C. McWilliams, 2009: Correction and Commentary for "Ocean Forecasting
in Terrain-Following Coordinates: Formulation and Skill Assessment of the Regional Ocean Modeling ...

shchepet

As of right now, ROMS expects to read the 3D-mode time step "dt" and mode-splitting ratio
"ndtfast" from the input script.

A potentially attractive alternative would be to specify "dt" and fast-time step "dtfast", and let
the model figure out the mode-splitting ratio within the code (along with a ...

shchepet

Hernan and Mark,

Are you serious about your idea of counting REAL-type numbers, like it is proposed in the latest fix,

my_StepTime=0.0_r8
MaxDT=MAXVAL(dt)

STEP_LOOP : DO WHILE (my_StepTime.le.(RunInterval+0.5_r8*MaxDT))

my_StepTime=my_StepTime+MaxDT

...

END DO STEP_LOOP

?

It is a bad ...

shchepet

I had a chance to examine the relevant KPP routines from a recent ROMS AGRIF,
files lmd_skpp.F and step3d_t.F of the following SVN versions
! $Id: lmd_skpp.F 726 2011-08-12 13:48:44Z gcambon $
! $Id: step3d_t.F 697 2011-04-11 12:35:17Z gcambon $
and found the following:

ghats(i,j,k), the same for ...

shchepet

Patrick,

First, the purpose of the above post was to fix what is obviously wrong -- mainly the discontinuity
of nonlocal flux at the base of PBL -- not to advance the physical accuracy of parameterization
(that would be a separate and a much longer subject), and yes, KPP is full of p -traps to ...

shchepet

Just to reply to the original question of John Warner.

Yes, KPP is constancy preserving in its NONLOCAL transport part, as well as everywhere else,
if you do it right.

Unfortunately, the original 1994 LMD paper buries this aspect into a rather sloppy notation making
it very hard to understand ...

shchepet

First, go to include subdirectory of your MPI directory three and inspect content of file "mpif.h"
to make sure that parameter MPI_PROC_NULL is present there [for MPICH versions 1.0.5p4 through 1.4.1p1
(current) it is just an integer parameter with value -1].

Secondly, this directory should also ...

shchepet

I vaguely remembered some flags such as MPDATA (or TS_MPDATA) could alleviate
such problems

This sounds like a fairly standard answer in ROMS comminity (and, in fact, in general
oceanographic comminity), however, it should be noted that contrary to a very popular
belief MPDATA is not a ...

shchepet

Nudging of Sea Surface Height (SSH) is a very bad idea and should not be used under any circumstances. The only way the model naturally accepts external SSH data is through lateral boundary conditions of the so-called Flather type.

The main point here is that the dynamical balance between SSH and ...

shchepet

Shchepetkin, 2000 (also cited in Ezer, Arango, Shchepetkin, 2002) refers to
http://www.atmos.ucla.edu/~alex/ROMS/pgf1A.ps
http://www.atmos.ucla.edu/~alex/ROMS/pgf1B.ps
which are odd and even pages of the same document (these files are designed
to generate a two-sided printout using one-sided ...

shchepet

And is it possible to plot two consecutive time steps to see how
the phase of zeta field changes from one 3D time step to the next?

shchepet

Yes, the problem here is due to the fact that the pressure gradient scheme described
in the 2003 paper WAS NEVER FULLY implemented in Rutgers ROMS.

That paper introduces the notion of adiabatic differences and harmonic averaging of
adiabatic differences as a method to enforce non-oscillatory ...

shchepet

With NetCDF 4.1.3 it is quite easy to specify what you want.

For example, if you want to use Intel Compiler, then you have to execute the following
sequence of commands (literally, just copy-paste and execute them one-by-one):

tcsh
setenv CC icc
setenv CXX icpc
setenv FC ifort
setenv F77 ifort ...

shchepet

how is it define baroclinic velocity: for practical purposes it is simply

u_bcl(i,j,k)=u(i,j,k,nstp)-ubar(i,j,nstp)
v_bcl(i,j,k)=v(i,j,k,nstp)-vbar(i,j,nstp)

where u,v_bcl(i,j,k) are baroclinic velocities;
u,v(i,j,k,nstp) are ROMS 3D velocities at certain time step as they occur in ROMS output ...

shchepet

This would be a very natural behavior of netCDF writing (and, as a matter of fact,
any buffered writing to the disk): as long as file is not closed or forcefully
synchronized, the data does not land onto the disk, but in the memory of the machine,
and consequently, is not accessible by any process ...

shchepet

...... -L/opt/gfortransoft/s_netcdf4/lib -lnetcdf -L/opt/gfortransoft/s_hdf5/lib -lhdf5_hl -lhdf5 -lz

resulted in

/home/rpacheco/PLOTROMS/src/crash.f90:1562: undefined reference to `nf_close_'
/home/rpacheco/PLOTROMS/src/crash.f90:1563: undefined reference to `nf_close_'
/home/rpacheco ...

shchepet

Pay attention to this:

/usr/bin/ld: Warning: size of symbol `mapcm1_' changed from 28 in libUTIL.a(hbackground.o) to 40 in /usr/local/lib/libncarg.a(mapbd.o)
/usr/bin/ld: Warning: size of symbol `mapcm2_' changed from 60 in libUTIL.a(cnt_frame.o) to 128 in /usr/local/lib/libncarg.a(mapbd.o)
/usr ...

shchepet

For the purpose of Ncview and my post above 2D-, 3D-, and 4D-variables means
2-, 3-, 4-dimensional arrays.

The picture you attaching is exactly what is expected: your netCDF file have shown
all four netCDF dimensions, namely xi_rho, eta_rho, s_w, and ocean_time for the
variable you are trying ...

shchepet

what level are they plotted at? maybe in the free surface?
No, not necessarily at surface or bottom.

Ncview plots exactly what you see on your screen at the time you hit its "Print" button.

Using main window of ncview you can navigate through variables present in your
file and select axes ...

shchepet

What is the input entry in your "ocean.in" file you are attempting to read when
the error occurs?

What compiler/version do you use?

The problem with the code

string=Vstring(is:ie)
LenS=LEN_TRIM(string)
READ (string(1:LenS),*) Rval(Nval)

is that the READ statement is not IOSTAT-protected ...

shchepet

Hernan, and all whoever may find it useful,

Some time ago I found that Computation of Courant number for the purpose of
diagnostics in done by averaging velocities toward the tracer point (see around
line 220 of diag.F),

my_Cu=0.5_r8*ABS(u(i,j,k,nstp)+u(i+1,j,k,nstp))* &
& dt(ng)*pm(i,j)
my_Cv ...

shchepet

All you have to do is to edit file "ROMS/Modules/mod_netcdf.F" around line 135 and change

!
! Netcdf file creation mode flag.
!
#if defined HDF5
integer :: CMODE = nf90_hdf5 ! NetCDF-4/HDF5 format file
#else
integer :: CMODE = nf90_clobber ! NetCDF classic format file
#endif

into

#if defined ...

shchepet

So if we do not define HDF5, the NetCDF file creation mode is "classic" and if you look at the unidata website, this option limits file sizes to 2Gb. Hence, if we run ROMS with a PERFECT_RESTART option and write-out the fields in double precision and our application is somewhat big, we can easily ...

shchepet

As far as I can tell, the answer is "no", you cannot save colorbar from ncview.

There are few controls about axes, title, etc... including option for none in each
category, that is to save plain plot of the field without any extra decorations,
axes, etc. This turns out to be the most useful ...

shchepet

This thread continues on viewtopic.php?t=2289
because the last post from Lyon was actually placed twice, here and following the link.

shchepet

NETCDF_ENDDEF - unable to end definition mode for file:
ocean_rst1.nc
call from: def_rst.F

Now it is something a bit new: it looks like it went through several netCDF calls
regarding creation of all dimensions and the variables, and then cannot finish its
definition. Do you have any other error ...

shchepet

Interestingly enough, in the netCDF web suite pointed by Lyon, among other
things it says

Should I start using the new 64-bit offset format for all my netCDF files?

No, we discourage users from making use of the new format unless they need it for
very large files. It may be some time until ...

shchepet

Official netCDF limitations for 64-bit format are practically unlimited, and are described here:

http://www.unidata.ucar.edu/software/netcdf/docs/netcdf/NetCDF-64-bit-Offset-Format-Limitations.html

i.e., there is still 4GByte limit on a fix-size variable and on record length
(aggregate size of ...

shchepet

So the distributed code had one version and now has a different, older version? It changed
without warning? And now we've built all these tools with the old version? Gotta love it.

Kate, I understand your frustration, which is, obviously an outcome of a communication
screw up on our side. But ...

shchepet

I do not object adhering with CF conventions in principle -- there is nothing wrong with them, but
I do not see how CF compliance changes this discussion, in sense that that I do not see what it
has to do with the way how the information about the specific type of vertical coordinate is stored
in a ...

shchepet

Jamie Pringle: ...I would prefer to know the "official" answers. File sniffing is easy to do poorly...

Actually it looks like you already provided the official answer yourself: it is what you call
"file sniffing".

The whole story of "new" coordinate emerged from the dissatisfaction with the ...

shchepet

Now it looks like this time your problem is different than before (when it was compiled
without -g -extra_flags ): now the code terminates immediately when attempts it to call
set_avg for the very first time, not when finalizing averaging.

Based on the fact that word 'hola' never gets printed in ...

shchepet

It appears that you are looking at very basic segmentation fault, which may be
associated with either ROMS itself, or to the particular netCDF version you are
using. set_avg_tile is a long routine. Most likely the breaking point occurs
inside

! Convert accumulated sums into time-averages, if ...

shchepet

HP.txt
Compiler flags : -heap-arrays -fp-model strict -openmp -fpp -ip -O3 -msse2 -free
Resolution, Grid 01: 0384x0176x030, Parallel Threads: 8, Tiling: 001x016
Compiler flags are not optimal: -heap-arrays ---> -no-heap-arrays , but you have to adjust stacksize limit.
Instruction set: -msse2 ...

shchepet

Kate, Hernan, Bárður, and whoever cares,

Here is a little test program to illustrate finiteness of cache lines -- the fact
that cache_line is more than just one number, as well as the pitfalls associated
with memory system architecture. It is worth spending 15 minutes playing with it.

All the ...

shchepet

People are going to more and more processors these days, not faster processors.
I've used about 100 at a time, but don't feel I can usefully use 1000. How do we get there?

Kate, Do you mean "cores" or "CPUs" or MPI nodes or hardware nodes
(motherboards)? We routinely use 256 cores "in house ...

shchepet

...This use of private storage to minimize memory requirements is fatal and
not possible in adjoint computations when the model is run backward in time...
Hernan, the horrendous practice in v.1.9 you are referring to is merely to
reuse the same scratch memory for two different purposes in a ...

shchepet

Do we spend our time staring at the code or waiting for the numbers to come
out? Do we want the fastest code that's perhaps more challenging to debug?

Kate, or perhaps anybody else: just compare step2d.F from v.3.4 vs. v.1.9 and
explain to me and to everybody else why v.3.4 is easier to debug ...

shchepet

DC(i,N(ng)-1)=DC(i,N(ng)-1)-CF(i,N(ng)-1)*sustr(i,j)/AK(i,N(ng)-1)

Yes, this is potentially a strong move, but it is also dangerous and can backfire.

What you are proposing is essentially to set the uppermost grid-box-side derivative
of the velocity field to du/dz = sustr/Akv at the free ...

shchepet

Regarding the performance comparison between ROMS 1.9 vs. 3.4: what
you are observing can be called as Fortran90 penalty . The fact that pre-Fortran 90
codes run faster than their supposedly more advanced successors actively using
new Fortran 90 features is nothing new, and it is actually noticed ...

shchepet

It must nevertheless be mentioned that the run-times reported above are single run
run-times and not averages over several runs, which would be the accurate procedure.

No, comparing averages is not the proper way to analyze it. Comparing the best against
the best would be more representative ...

shchepet

....
Maximum barotropic Courant Number = 9.27620490E-01
....

is a bit too much. Increase ndtfast slightly, by 10...20%,
or reduce main time step by the same fraction.

Also make output at every time step to see how the instability
develops.

shchepet

...So what? The best results for the SOLITON problem are:
for ROMS 1.9

23.0 sec using 6 threads, 12x32 tiling

vs. for the new code

32.8 sec also 6 threads, 12x32 tiling,
Compiler flags: -pc80 -xSSE4.2 -auto -stack_temps -openmp -fpp -ip -O3 -free

The above slightly degrades to

35.6 sec ...

shchepet

/opt/intel/Compiler/11.1/072/lib/intel64/for_main.o: In function `main':
/export/users/nbtester/efi2linux_nightly/branch-11_1/20100415_000000/libdev/frtl/src/libfor/for_main.c:(.text+0x38): undefined reference to `MAIN__'

This is caused by an attempt to compile code without compiling mpc first ...

shchepet

.....
analytical.F:446: error: missing binary operator before token "ISWAKE"
.....

There is a type on line 446: It says

#if defined ANA_FSOBC && !defines ISWAKE

should be

#if defined ANA_FSOBC && !defined ISWAKE

i.e., !define s --> define d

On my machines it still compiles and runs ...

shchepet

Answer to 2: Running with FFLAGS := -no-heap-arrays -fp-model precise -openmp -fpp -ip -O3 - axSSE4.2 -free and increased stack size (16384) using same tilling (6 x 32) on :
12 treads the time was: 39 sec.
8 treads the time was: 44 sec.

This is a good news: looks like hyperthreading works now ...

shchepet

I usually run current version of ROMS using the build.bash script, where netcdf is located by the following lines:
export NETCDF_INCDIR=/usr/local/include
export NETCDF_LIBDIR=/usr/local/lib

Then you should set

CPP = /lib/cpp -traditional -D_OPENMP -D__IFC -I/usr/local/include

LCDF = -L/usr ...

shchepet

...and relative to where I began the simulations are running much faster now.

The point is that you know exactly where you have started, but you do not know where are
you heading in sense that you do not know when it is good enough to stop. And I feel that
it will take several iterations from ...

shchepet

So what was the problem?

Was it just because changing the brand of MPI?

If so, what kind of MPI you was using before?

Any specific setting on tunable parameters?

shchepet

So as long as ROMS runs with default stack size there is no point in changing it right?
In Ubuntu/bash I can increase the stack size by setting e.g. ‘ulimit -s 16384’.

This time you actually made me do my homework because what you are reporting --
essentially insensitivity of your code ...

shchepet

This sounds more like MPI problem, not directly ROMS problem, although
it is triggered by ROMS pushing MPI to the limit.
............
0 00:00:00 0.000000E+00 1.963001E+04 1.963001E+04 1.659940E+17
DEF_HIS - creating history file: ./output/ocean_his_0001.nc
rank 447 in job 1 n104_34487 caused ...

shchepet

Did you create directory called "output" within your working/scratch
directory where you running your job?
.....
STEP Day HH:MM:SS KINETIC_ENRG POTEN_ENRG TOTAL_ENRG NET_VOLUME

0 0 00:00:00 0.000000E+00 1.963001E+04 1.963001E+04 1.659940E+17
DEF_HIS - creating history file: ./output/ocean_his ...

shchepet

...Dell ... Xeon X5560 - 2.80GHz .... What should I do ...
1) A processor with higher clock?
2) Or increase the number of nodes?

Absolutely increase the number of nodes and absolutely stay away from Dell.

Your machine is already too expensive and any further increase of clock speed would bring ...

shchepet

First, it is a common experience with such a problems that they may be a bit
tricky to start up. How did you create initial conditions? Possibly some kind of
Matlab script. Because your kinetic energy starts with zero, it looks like you
have created temperature and salinity field for the initial ...

shchepet

Also caught my attention is

leommcruz
cpu: Intel(R) Xeon(R) CPU E5410 @ 2.33GHz cache size : 6144 KB
I'm running a grid with 200x256x30 ...With a 1x8 tile I got the best openmp result ...

and

leommcruz
Regarding the tile option, in my case the best results I got were using 2x2 tile (my ...

shchepet

the cpu time should be the sum of all the cpu-times together
since a parallel run is never 1/P as fast as a sequential run (in SMPs),
the sum of cpu time will be higher.

ROMS reports two times: CPU time spent and wall-clock time. This is the intent,
Wall clock time is easy to understand. CPU ...

shchepet

So you recover some of the performance loss, but there is still a way to go.

QUESTION: where does the 8x81 tiling come from, memory configurations?

I made it up just as a first guess. Generally for these kind of problems one wants to chose the size of tile to reach the best possible compromise ...

shchepet

Grid size: 1398 x 726 x 1
Tilling: is 3 x 2

This is not the best way to run ROMS.

Try to set tiling to 8 x 81 (yes, meaning eight by eighty one),
rerun it and report your finding back to this forum. We will
continue after that...

Motherboard: ASUS P6T SE, X58 chipset
RAM: DDR3 ...

shchepet

I just saw this thread of conversation and to my amusement found what I would characterize as
pristine naiveness: did not we been through this before?

MPI vs. OpenMP .... influence of I/O, and, ... the usual award winning phrase that "benchmarking ROMS is not trivial..."

Contrary to popular ...

shchepet

Lucky you, -- you have system administrator.

Well, these are colors ... so they are just a matter of taste.

"bright" is general purpose and works for local SST. On a large scale model, that is the whole Pacific
"banded" can show more detail, while keeping wide range; "rainbow" is overused here in ...

shchepet

This post is for people who like to use "ncview" as a quick tool to visualize
netCDF files, such as ROMS output (or of any other origin), and who wants to
customize color palettes built-in into by this tool.

Ncview is designed by David Pierce,
http://meteora.ucsd.edu/~pierce/home_page.html
and it ...

shchepet

What I want to do is I want to store 2d variables (zeta, ubar,vbar)
frequently (say hourly or daily) without 3d variables (temperature,
salinity, u, v) on history file. And I want to store 3d vaiable with
longer time step(say 10day or month) on average file.

This is possible "as is" with ...

shchepet

Hernan,

According to netCDF manual nf90_64bit_offset is additive to the primary netCDF
flag, so it makes sense to use

ierr=nf_create(ncname, nf_clobber+nf_64bit_offset, ncid)

when you want to create file with nf_clobber status (i.e., overwrite whatever file
may already exist) AND with 64bit ...

shchepet

No, Mark, it is much simpler that that: you do not have to go to netCDF 4, although you can.
Version 3.6.3 -- the latest and final version of 3rd generation of netCDF is perfectly capable
to by-pass the 2GByte limitation.

All you have to do is to change

ierr=nf_create(ncname, nf_clobber, ncid ...

shchepet

Yes, what you did is correct, and what it was before is wrong,
and, in fact, is a parallel bug if the code is compiled/used in
Open MP mode.

The explanation is as follows: if you examine

./roms-3.1/ROMS/Functionals/ana_grid.h

You will find that shared arrays xr(i,j)=... and yr(i,j)=... are ...

shchepet

./oceanO < ROMS/External/ocean_esh.in
./oceanO: error while loading shared libraries: libiomp5.so: cannot open shared object file: No such file or directory

This is easy to fix: libiomp5.so belongs to intel compiler, and is
located in /opt/intel/Compiler/11.1/056/lib/intel64 for the compiler ...

shchepet

rm -f -r /usr/local/Compilers/make_macros.mk
cd /usr/local/Build; -c -heap-arrays -fp-model precise -openmp -fpp -ip -O3 -xW mod_kinds.f90
/bin/sh: -c: command not found

from Kate

The command not found is your compiler. What happens when you type "which ifort"?

No, the error message says ...

shchepet

What is in your .cshrc file in home directory of the machine you use?

A typical practice in supercomputer centers to install a new version of the compiler
when it becomes available, but keep the old one, for just in case. This is also what
we are doing here at home. As the result, at any given time ...

shchepet

Intels keeps modifying both compiler and compiler usage, along with hardware, so compiler
flags are generally chanhing all the time.

Compiler setting and flags like

Compiler command : ifort -fpp2 -openmp -pc80 -axP -align dcommon -auto -stack_temp

Compiler flags : -O3 -IPF_fma -ip -xW -free

are ...

shchepet

1. is Akv the classic eddy viscosity or is it VISC?

No. VISC is horizontal viscosity, sometimes known as "eddy viscosity", basically parameterization of horizontal momentum exchange by interaction of eddies smaller than horizontal grid size. In practice it is chosen most of the time to ensure ...

shchepet

Several years ago I have tested a FORTRAN compiler called PathScale and compared
it against the usual suspects -- ifort and pgi. What I found is that when using
intel CPUs (Xeon, Pentium 4 at that time) ifort was definitely superior to both
alternatives; when using Opteron CPUs (these were older ...

shchepet

Most likely your problem is related not to the ROMS code you use,
but to the machine and/or its operating system.

Open MP programs can be executed in a single processor machine, but,
of course, threads go sequentially from one synchronization barrier
to the next, resulting in no parallel gain.

To ...

shchepet

OK, I is kind of serious, since it affects several of our codes.
I investigated it a little bit further, and I found that:

1. the failure to return non-zero error status depends on the
initial symbol of the string. To see it, compile and execute
the following code:

program reading_error ...

shchepet

I encountered the following behavior:

internal Fortran reading operation fails to return non-zero status if
one attempts to read a real*8 from a string which is does not contain
a valid number, and therefore the reading operation must fail.

This may have some implication for ROMS, ever more for ...

shchepet

The difference between the 2003 paper and what you see in the code is explained by the fact
that equations in the paper are written assuming that density "rho" is the whole density (meaning
that it is about 1030. kg/m^3), while the code is written assuming that density "rho" is density ...

shchepet

It depends upon which code are you using. But it easy to track it down
in all three cases.

Let's Check it out.

Rutgers (say. v. 3.1): file pre_step3d.F: array "swdk" is assigned
its value at/near line 345, and it is a positive-definite field. Then it
is used at about line 784, an now observe the ...

shchepet

There is a bit of logical inconsistency within ROMS convention, but as the result, it makes it more
in line with oceanographic tradition, and somewhat with common sense.

The rules are:

--- Positive direction of vertical axis Z is upward.

--- Model INTERNAL fluxes are positive if they are pointing ...

shchepet

Then you should use Makedefs.generic as Makedefs, or make your own,
depending on what type of computer, operating system, and FORTRAN/C
compilers you have.

I usually supply about 10...15 Makedefs.XXXX files, which are for
different machines and advise people how to create their own: basically
use a ...

shchepet

The "ncrename" I was was referring to belongs to a collection of several
basic "tools" designed to manipulate netCDF files (look at target/macro
tools/TOOLS in your Makefile for AGRIF ROMS; in order to use them, you must
compile them first, hence type "make tools" command; this creates several ...

shchepet

NRREC in keyword-block "initial" specifies record number within the
file from which you want to start simulation, say

initial: NRREC filename
2
rst.0000.nc

means that you have have file "rst.0000.nc" which contains several
records, and and you want to use record 2 as the initial condition ...

shchepet

F.Daryabor,

Based on the format of your roms.in file, and your working habits
(e.g. subdirectory named ROMS_FILES) you are using AGRIF ROMS.
This code has capability to arrange your output into sequence of
netCDF files, rather than single file with unlimited dimension.
This way you can avoid ...

shchepet

Beckmann-Haidvogel "rx0" number is relevant only because it controls Haney number "rx1" (under all other conditions -- number of vertical levels and stretching -- other kept the same, the two are proportional to each other), and also because you know Haney number only after you setup your vertical ...

shchepet

But what about Ekman spiral? If you solve the problem analytically, then you find that water velocity vector always starts at 45 degrees to the wind stress applied. That applies to both classical Ekman problem as well as turbulent. Then, once you start subtracting something like 0.5*(u(i,j,N)+u(i+1 ...

shchepet

It reminds me that we, as a community, should come up with an
identification standard for different types of vertical coordinates
in order for pre- and post-processing software to recognize it
correctly and automatically.

I personally no longer use the original S-coordinate for quite
a while, more ...

shchepet

It should not dissipate, if that number is 0.47. However, you specified the whole length
of your channel as "wavelenght_of_interest". I suspect that the actual wave is shorter
than that, so it may be not 0.47, but is more. As of right now, your signal travels 7.5km
per time step, and it takes about ...

shchepet

What is your time step "dt" (for 3D mode) and mod splitting ratio "ndtfast", and maximum
barotropic Courant number "Cu_max" as it is reported by the model standard output?

Also try to estimate wtat is the ratio of

2*pi * dt * wave_phase_speed / wave_length_of_interest ?

where pi=3.141596 ...

shchepet

The difference between "finite differences" and "finite volume" is saddle, but significant.

The main thing is interpretation of the gridded data: "finite-volume" assumes that u(i,j,k) is AVERAGE of field "u" over control volume dV(i,j,k), while finite difference assumes that it is instantaneous ...

shchepet

The rationale here is that some CPP-switches are project-dependent and are to be defined for a specific physical configuration according to its needs. They are usually set by cppdefs.h (or what ever small configurational include files in roms-3.0) .

But there is another group of CPP-switches which ...

shchepet

There is no emphasis for description of the particular main 3D time
stepping procedure used in ROMS v. 1.8....2.2 in the Shchepetkin
McWilliams,2005 ROMS paper because this algorithm was always considered
as a provisional variant to be replaced with a more refined one.

The main 3D procedure of v. 1 ...

shchepet

Steve,

I did not see your message until recently. Regarding my post about establishing 1CPU reference point is to evaluate the performance of "internal code", which is consistently overlooked. Not only in ROMS discussion on this board, but in a wider community as well: people tend to be happy with ...

shchepet

There is a permanent solution to this problem practiced in UCLA and Agrif codes for the last 7+ years: create a mechanism for automatic tracking of the status of each CPP switch in "cppdefs.h" by automatically generating an analog of Rutgers "checkdefs.F" (known as "check_switches1.F" in UCLA and ...

shchepet

The remarks like "LOOP WAS VECTORIZED" are actually a good sign. It means that compiler
is able to optimize the code in some situation. Ideally every innermost loop should be
vectorized.

[ in Intel's terminology "vectorized" means that consecutive 8-byte real numbers are
loaded in quads, that is ...

shchepet

At first, I suggest you compile netCDF library and the code using GCC compilers, because now you have to solve double problem: you have code-specific problems and you have Intel specific problems as well (for example, you say that you are not able to install 64-bit version of Intel compiler on you ...

shchepet

First, Intel Core 2 dual core processors is a good choice, by far the best computing power per unit price. However, to utilize this fully, you must use Intel IFC compiler: no substitution. You can download a free non-commercial version of it from Intel's web suite

http://www.intel.com/cd/software ...

shchepet

Yes, it is better to fix the problem by controlling your environment rather that hardcoding parth inside FORTRAN code, so you are better off by setting

-I/directory/of/netcdf/include/lile

among you CPP flags, and

-L/directory/of/netcdf/library -lnetcdf

among your linker flags, but this ...

shchepet

pmask settings have any effect only if viscosity is nonzero.

The code may be run with or without explicit viscosity, if the upstream-biased advection is
used, because the advection scheme is already dissipative enough on grid scale to avoid
numerical issues.

If, on the other hand, centered ...

shchepet

As a side remark here is that setting pmask=2 is acceptable along straight boundaries
to account for the fact that tangential velocity is located half-grid interval from the boundary,
and therefore

pmask(i,j)*(u(i,j,k) - u(i,j-1,k))

accurate evaluates derivarive near the boundary, if one of ...

shchepet

We have several Tyan Thunder K8W (S2885) in our lab.

Although frequently advertized as "DUAL CORE READY", this board MAY or MAY NOT
accept dual-core CPUs, depending on revision number. So before you decide to upgrade
CPUs get your flashlight and make sure that you can read revision number

04MOAb ...

shchepet

I am running both WindowsXP (whatever come with the Laptop made by Dell) and Linux
Mandriva 2007 with 2.6.17-8mdv kernel. This kernel is SMP cabable: with the release of
2007 operating system Mandrake (now Mandriva) makes ints "normal" kernels be SMP,
thus abolishing the distinction between SMP ...

shchepet

MPI is useless in this type of configuration, predominantly because of slow and shared
memory bus, see

http://www.atmos.ucla.edu/~alex/ROMS/poorman_revisited.pdf

On the other hand, OpenMP works remarkably well and I can get perfect scaling,
and overall excellent performance: my CoreDuo 1.83GHz ...

shchepet

...There is an also easy-to-use, compule once --> use forever FORTRAN code which does
generic extraction of perimeter from ROMS climatology file,

http://www.atmos.ucla.edu/~alex/ROMS/perim.F

Just compile it, name executable "perim" and save it into your ~/bin directory, and then say

perim WS ...

shchepet

...something else caught my attention. It looks like there is a missing factor of 1/2
in KE computation, since

KE = intergal of V^2/2

and not just of V^2. The factor of 0.5 in ke2d computation comes from averaging,
0.5*(u(i+1,j)^2+u(i,j)^2) approximates u^2 and not u^2/2. So, unless ke2d is ...

shchepet

There is an error here: multiplier *(rho(i,j,k)+1000.0_r8) should not be present in ke2d calculation. Just change it

ke2d(i,j) = ke2d(i,j) + Hz(i,j,k*(0.5_r8*(u(i ,j,k,nstp)*u(i,j,k,nstp)+....

You may also look at diag.F routine from

http://www.atmos.ucla.edu/~alex/ROMS/roms.tar

to use it for ...

shchepet

Dear Kate, Rich, and all,

I am just wondering: sometime ago, back in February 2004 Hernan asked me to revise mex-function interface and fortran code "mexsepeli" in order to enable it to work on Linux platform and being called from Matlab on Linux. Which I succeeded to do. I knew that it was needed ...

shchepet

Regarding which code to use: the short answer is use a version which is the most suitable for your particular needs. This is because there are many peripheral developments and submodels driving my applications for which you (and most people except those directly involved) probably do not care about ...

shchepet

Actually what is proposed by Hernan is not consistent with Lilly discretization on a C-grid, even thought it may be viewed as approximatrely correct.

Hernan's procedure implies that velocity components are averaged from their U-,V-points to RHO-points first, and then squered there. This is a WRONG ...

shchepet

...as far as I know, the compressibility switch has never been implemented in Rutgers codes, but is now standard in UCLA codes (SPLIT_EOS is #defined all the time).

...and yes, it leads to additional pressure gradient errors, if EOS does not allow you to distinguish between in-situ and adiabatic ...

shchepet

This reply is not a to be understood as statement of policy, but rather explanation of de-facto settled practices.

If no boundary conditions are specified, then it means that they are CLOSED everywhere. It means: (1) no normal flow; (2) Neumann (zero-gradient extrapolation rule) for T,S; and (3 ...

shchepet

At first, I consider it a bad habit, because it can be avoided by arranging your output
into sequence of files, rather than single big file. Secondly, it restricts portability: you
must have an operating system which has Large File Support and you can no longer
open it from a 32-bit Matlab, and ...

shchepet

We routinely run this code on FC4 as well as on other flavors of Linux.

Furthermore, from user's point of view in makes no difference, whether it it FC4 or Redhat 9, or Enterprise Linux or Mandrale 9.0, 9.1, 9.2, 10.0. 10.1, 10.2/200LE or 2006.

All you have to have is:

Intel Fortran compiler ...

shchepet

No, changes of this kind will not cause any conflict.

All ROMS codes are written in such a way that any physical code interacts only with arrays z_r and z_w and makes no assumption about where these arrays are coming from. In fact, the only places in the code which are aware of the S-coordinate are ...

shchepet

The second aspect of this problem --- interference between computation of 3D--> baroclinic mode forcing term and the restart of barotropic mode --- is a more serious issue because all ROMS codes (UCLA, Rutgers, Agrif) are affected by this issue.

The problem comes from the fact that the full 3D ...

shchepet

This observation is generally correct: the timestepping algorithm of barotropic modes of both Agrif and Rutgers code (all versions of both) cannot be accurately mapped into Eqs. (2.27)---(2.30) of the time stepping paper because r.h.s. terms saved from previous time step contain barotropic pressure ...

shchepet

I forgot to mention: because on my machine(s) I always have multiple compilers and multiple operating systems, I always place netCDF library into the "lib" directory of the compiler which was used to compile it.

This way, I just have to specify

LCDF = -lnetcdf

without path to it. The path is ...

shchepet

I start my reply with nganju's question about light speed-up (~1-2%) when hyperthreading is turned off.

Yes, this is consistent with my own experience and you should always turn hyperthreading OFF in BIOS of your dual-Xeon machine. I observed improvement by as much as 7...10%. In the case of ...

shchepet

Dear Mark, and everybody,

The problem, as already pointed out, is related with stack size limitation and may
be fixed in many cases if one can increase or even unlimit stacksize. This, however,
is not as innocent as it might sound.

In older days (ROMS 1.9 and earlier) scratch arrays were pre ...

shchepet

Dear All,

First remark is that this set_weight modification is not a bug fix, but it would better fall into category of optimization of code usage. The purpose of filter is to remove fast frequences of the barotropic motions to prevent their aliasing and instability of the model. The downside of it ...

shchepet

It is known that the power function weights for time filtering of the vertically integrated fields ( POWER_LAW CPP option) result in unstable code when the splitting ratio ( ndtfast ) becomes too small. For a natural setting of ndtfast=60 to ndtfast=70 (most of the realistic basin-scale problems ...

shchepet

Dear All,

I was well aware of this feature of SPEM/SCRUM/ROMS codes for years (at least 5 years) and do not consider this feature as a bug. Neither I consider the latest fix as an urgent measure to be done immediately, and I do not see any reason for panic or statements like I cannot believe that ...

shchepet

Atmospheric pressure forcing can be easily included into the pressure gradient routine without
messing with body-force-type algorithm. As the matter of fact, any pressure gradient force
calculation algorithm involves (explicitly or implicitly) vertical integration of density field from
surface to ...

shchepet

I glanced the whole thread of conversation and found it quite disapointing, in sense that there is virtually no attempt to make a meaningful interpretation of the results.

Because of split-explicit time stepping in ROMS code (all of them), most MPI-related latencies occur in 2D mode, where many ...

shchepet

I found it successful and easy to install too (I am using Mandrake Linux 9.2 and 10.0). However performance is significantly lower that that of Intel IFC compiler. For a test problem I am getting 45 seconds run time while the same test, but using Intel compiler takes only 28 seconds and 36, if PGF ...

shchepet

As you probably know, I made some revision of KPP codes in ROMS last summer (2003) with the view to (1) update it first to the state of NCAR model of Bill Large of 2003 as much as I can; and (2) move on with some new ideas about handling of Ekman Boundary Layers and other things, like Monin-Obukhow ...

shchepet

Dear All,

The main difference is that Rutgers ROMS (all versions 1.x...2.0/TOMS) use Adams-Bashforth 3 time step for 3D momentum equation (inherited from SCRUM 3.0 code) and LF-TR predictor-corrector (PC) algorithm for tracers. The barotropic mode uses LF-AM3 PC algorithm with some forward-backward ...

shchepet

Hernan, and All,

I think the idea to abandon the ROMS e-mail list in favor of bulletin board is not a good one: the discussion actually died out.

I understand that this was done because of hacker attacks who sent viruses, but in reality we just yield to hackers. Frankly, these viruses affect only ...

Search found 188 matches