NETCDF error

[martinspenke]

Dear Daniele,

From time to time i obtain NETCDF error in yambo_3.4.2 :

Code: Select all

P001: [ERROR] STOP signal received while in :[06.01] G0W0 on the real axis
P001: [ERROR][NetCDF] NetCDF: One or more variable sizes violate format constraints

when i run heavy calculations. Is there any possibility not to compile with netcdf ?

Best wishes
Martin

[Daniele Varsano]

Dear Martin,
If I remember well it should be possible, you can try.
Anyway this kind of error it is not a netcdf problem but a fingerprint of something else going wrong.
In any case, 3.4.2 it is not supported anymore, so I suggest you to download 3.4.3 or 4.0.x.
If the problem persists you can post the input/report etc. and we will have a look.

Best,

Daniele

[martinspenke]

Dear Daniele,

If I remember well it should be possible, you can try.

How can i disable netcdf support ? I remember you told me some while ago netcdf for yambo_3.4.2 is mandatory.

In any case, 3.4.2 it is not supported anymore, so I suggest you to download 3.4.3 or 4.0.x.

yesterday i downloaded the latest revision of yambo_3.4.2 from svn ([VER] 3.4.2 r.5108).
I see nothing with yambo_3.4.3

Bests
Martin

[Daniele Varsano]

Dear Martin,
sorry I made some confusion. As you observed the stable version is 3.4.2.
And yes, the netcdf are mandatory, the difference with 4.0.x. is that in the latter they are automatically downloaded by the Makefile.

As I indicated in the last post, usually the netcdf error are fingerprint of other problem and it is not sure that it is related with netcdf, even if this kind of error may appear when netcdf variables exceed a defined limit. Can you please post your input/report/log files?
If this is the case may be you can overcome it by using fragmentations or avoiding I/O. All this workaround anyway are much more flexible with the 4.0.x

Best,
Daniele

[martinspenke]

Dear Daniele,

Unfortunately i deleted my report and log files. I will reproduce the error and will come back.
I actually fragmented wave functions by p2y -S N.
what is keyword for avoiding I/O ?

As of 4.0.x, it is unfortunately very slow in comparison to yambo_3.4.2 on cluster. Thus, i prefer to use the traditional stable yambo_3.4.2.

I also have another problem :
I obtain these warnings when i perform a frequency dependent GW calculation on top of COHSEX energies :

Code: Select all

<02s> [WARNING] QP corrections transferred in energy levels duplication
<02s> [WARNING]Global and response function energy levels have different metallic character

I do not understand the first warning.
And for the second warning, i do not have a metallic system. my system is an insulator with a direct band gap of 0.5 eV.
I do not obtain these warning in yambo_3.4.1 for instance.

Bests
Martin

[martinspenke]

Dear Daniele,

Please find attached the necessary files.
The input file was generated by yambo_3.4.1.

Best wishes
Martin

[Daniele Varsano]

Dear Martin,

It is not obvious what is going on. Most probably too much variables in the database to be written or a too large file (> 2Gb).
Which is the last database written in the ./SAVE directory?


My suggestion without any particular order are:
1) Try to see if the -S option fragment the database; ---> Edited: please discard this and see post below

Code: Select all

yambo -S

2) Fragment your calculation by hand: i.e instead of:

Code: Select all

%QPkrange                    # [GW] QP generalized Kpoint/Band indices
  1|   73|  14|  30|         
%

More runs for instances 14-20 ; 20-25 ; 25-30

3) Recompile the netcdf enabling the large size support

Code: Select all

  --enable-netcdf-LFS     Activate NetCDF Large File Support. Default is no.

when configuring

4) Switch to yambo 4.0, if the parallelism variable are set properly it should not be slower.

Hope this help,

Best,
Daniele

[Daniele Varsano]

Let me add something about fragmentation.
I was thinking it was a feature of 4.x, but is already in 3.4:

When building up your inout files add the

Code: Select all

-V io

options.
You will see in input a new keyword:

Code: Select all

DBsIOoff= "none"             # [IO] Space-separated list of DB with NO I/O. DB=(DIP,X,HF,COLLs,J,GF,CARRIERs,W,SC,BS,ALL)
DBsFRAGpm= "none"            # [IO] Space-separated list of +DB to be FRAG and -DB NOT to be FRAG. DB=(DIP,X,W,HF,COLLS,K,BS,QINDX,

You can try to set: "+W" in DBsFRAGpm in order to fragment it. Or "W" in DBsIOoff in order to avoid to write it on disk.
I would try this first.

Of course check if it is ndb.W the database that is cauing problems.

Best,
Daniele

[martinspenke]

Dear Daniele,

many thanks for your helps.
However, when i use -V io in yambo_3.4.2, i do not obtain :

Code: Select all

DBsIOoff= "none"             # [IO] Space-separated list of DB with NO I/O. DB=(DIP,X,HF,COLLs,J,GF,CARRIERs,W,SC,BS,ALL)
DBsFRAGpm= "none"                # [IO] Space-separated list of +DB to be FRAG and -DB NOT to be FRAG. DB=(DIP,X,W,HF,COLLS,K,BS,QINDX,

-V io is a feature of yambo_4.0.x.

The problem with netcdf is the large ndb.W file (>2 GB).
I also compiled yambo_3.4.2 with --enable-netcdf-LFS but it did not help.

Is there any way to tell yambo_3.4.2 not to print out ndb.W file ?

I switched to yambo_4.0.4. However, i wish i could do it with yambo_3.4.2, too.
Is the support for yambo_3.4.2 terminated ?

Best wishes
Martin

[Daniele Varsano]

Dear Martin,
3.4.2 it is still supported. You are totally right -V io is a feature of 4.0x. Here in Rome it is very hot and I made again a mess.
Unfortunately, or fortunately for me I will be not fully connected for the next two weeks. Hopefully someone else will help you.

Best,
Daniele