Hi all,
How can I append the new avg ouput when I restart the model to the last avg.nc file ?
With regards,
Francis.
how to append?
Re: how to append?
You can concatenate ROMS (and other) netcdf files with an unlimited, or 'record' dimension, using the ncrcat tool in the NCO toolbox
http://nco.sourceforge.net/
http://nco.sourceforge.net/
John Wilkin: DMCS Rutgers University
71 Dudley Rd, New Brunswick, NJ 08901-8521, USA. ph: 609-630-0559 jwilkin@rutgers.edu
71 Dudley Rd, New Brunswick, NJ 08901-8521, USA. ph: 609-630-0559 jwilkin@rutgers.edu
-
- Posts: 21
- Joined: Wed Jun 13, 2007 1:22 pm
- Location: Indian National Center for Ocean Information Servi
Re: how to append?
Can't we do in ROMS?
One more answer I want to know is, how we have to give NtileI, NtileJ options in .in file. Actually we are having 40 nodes(80 processors), my domain size is 109*62*40, so what is the correct choice for that?
When I gave NtileI=2, NtileJ=16, it took 3min/day, from there onwards if I increase the no. of processors also, the time is not decreasing but in contradiction time is increasing with processors (whatever choice of NtileI, NtileJ).
So is there any threshold for MPI jobs?
Best regards,
Francis.
One more answer I want to know is, how we have to give NtileI, NtileJ options in .in file. Actually we are having 40 nodes(80 processors), my domain size is 109*62*40, so what is the correct choice for that?
When I gave NtileI=2, NtileJ=16, it took 3min/day, from there onwards if I increase the no. of processors also, the time is not decreasing but in contradiction time is increasing with processors (whatever choice of NtileI, NtileJ).
So is there any threshold for MPI jobs?
Best regards,
Francis.
Re: how to append?
Yes, you can do it in ROMS. There is a flag LDEFOUT in the input file. Set this to F and ROMS will append to the average and history files.
For your parallel performance, 109x62 is a rather small grid. It also has very few factors of two in the Lm,Mm values. You can run some tests with the BENCHMARK setup which was designed for testing such things. It comes in three different sizes, so you can explore how much better your parallel performance is for large problems compared to small ones. Even BENCHMARK1 is bigger than your grid and it is noticeably worse than BENCHMARK3 for scaling to tens of processors.
For your parallel performance, 109x62 is a rather small grid. It also has very few factors of two in the Lm,Mm values. You can run some tests with the BENCHMARK setup which was designed for testing such things. It comes in three different sizes, so you can explore how much better your parallel performance is for large problems compared to small ones. Even BENCHMARK1 is bigger than your grid and it is noticeably worse than BENCHMARK3 for scaling to tens of processors.
-
- Posts: 21
- Joined: Wed Jun 13, 2007 1:22 pm
- Location: Indian National Center for Ocean Information Servi
Re: how to append?
U r right Kate and thanks for ur reply.
I could do with turning on LDEFOUT in .in file.
When I make my grid to 110*62*40, it taking 2min 20 sec for one day with 22processors and the old grid took the same time with 32 processors. Thanks once again for ur reply.
I could do with turning on LDEFOUT in .in file.
When I make my grid to 110*62*40, it taking 2min 20 sec for one day with 22processors and the old grid took the same time with 32 processors. Thanks once again for ur reply.