xconv crashing during conversion

[a.elvidge]

Hi,

During conversion of pp files to a netcdf file xconv crashes. Because of the large size of my .nc file I am using a beta version (I believe) of xconv here (/home/n02/n02/jwc/um/xconv/xconv1.92) to enable 64 bit offset .nc output format.
I have used this plenty before for exactly the same kind of jobs without problem. I have tried deleting the file and retrying and have notices that it appears to crash when the new .nc file reaches a certain size (about 9gb). Could this be simply a disk quota issue? On SAFE, it appears that my personal usage is fine but that the n02 group usage may be at or over its limit?

Thanks, Andy

[grenville]

Andy

It's not a disc quota problem. Please let us know which file(s) are causing the problem.

Grenville

[a.elvidge]

Hi Grenville,

I'm trying to convert all the xfxkla_pa* files to 060106_1.5km.nc in /work/n02/n02/aelvidge/xfxkq_060106

I've just tried once more with the same result… the file reaches about 9gb then xconv (/home/n02/n02/jwc/um/xconv/xconv1.92) crashes (disappears).

Thanks, Andy

[a.elvidge]

Jeff,

I have used xconv1.92 on exactly the same stash files (same job) but for different runs. It's suddenly stopped working. It crashes and a 'killed' message appears on the hector command line. It is not just failing on this one particular job; I just unsuccessfully tried it on a set of UM output files I had previously (about 2 weeks ago) successfully converted to 64-bit offset.

Thanks, Andy

[jeff]

Hi Andy

You shouldn't be using /home/n02/n02/jwc/um/xconv/xconv1.92, but it hasn't changed for almost 2 years so I don't know why it stopped working. Try using /home/n02/n02/jwc/bin/xconv1.92 instead and see if that works.

Jeff.

[a.elvidge]

Hi Jeff,

You advised I try /home/n02/n02/jwc/um/xconv/xconv1.92 in response to a previous query I raised on here, and I have been using it ever since. Unfortunately /home/n02/n02/jwc/bin/xconv1.92 is not working for me either - exactly the same problem. I have also tried a colleague's computer (who is on Linux, I am on Windows using XMing), and the problem still occurs.

Perhaps an idea might be if you could try using it for a similar job, or perhaps copying across my files to try?… What ever you think would the best way to investigate the problem further.

Thanks, Andy

[a.elvidge]

I copied xconv1.92 across to the uea cluster along with my UM output files, and the conversion works. So it appears to me likely to be either a problem with hector or with my account in particular. Its as if the program times out and is forced to close after it has been running for a certain time (but only when running a job… when left idle it stays open).

Andy

[grenville]

Andy

Try running xconv on the larger memory server (ssh -Y lms from a HECToR login node) (​http://www.hector.ac.uk/support/documentation/guides/lms/ for details). You should be able run interactively for a while on that machine.

You will need to apply for an account on the lms through SAFE, but this should be quick.

Grenville

[a.elvidge]

Hi Grenville,

Thanks, this works. I tried again without logging into lms and it is still not working. So it would appear that, at least in my case, running xconv in lms is now a requirement for large conversions.

Thanks, Andy

[jeff]

Hi Andy

Its not a xconv problem but something they must have changed on hector. You could try asking the hector helpdesk, but I suspect they will say you shouldn't be doing this sort of processing on a hector login node. This is what the lms is for so using it is definitely the best idea.

Jeff.