Skip navigation
NASA Logo, National Aeronautics and Space Administration
Currently Being Moderated

GEOS5-DAS failures on discover

VERSION 1 
Created on: Nov 22, 2008 8:38 AM by Carlos Cruz - Last Modified:  Nov 22, 2008 8:59 AM by Carlos Cruz

These are som problems experienced by a user running the GEOS5-DAS on DISCOVER:

 

1. A few of my jobs failed because they could not complete "acquire" observations from the mass storage on time. DAS submits a separate job to bring over obs and gives the job 1hour wall time. And it's not always enough. It never happened on Palm. So, the relationship between the "discover" and the archiving system is not good enough.

( I finally found in the DAS scripts where the time is set and doubled it. It may help.)

 

2. The absence of large Scratch space  forces  users to use  nobackup system  as temporary space. I keep running out of it after having asked to double it twice. The thing is that currently we run the DAS at .5X.5 resolution, so all the temporary output files are huge, and the jobs fail before archiving the output on the mass storage system.

If a job fails for any reason it does not clean after itself, used space keeps accumulating and subsequent jobs fail. Unless there is a lot of babysitting.

As you can see, these problems can be resolved (maybe!) but it seems to me the system as not setup optimally at least for GEOS DAS users. So, there is no complex issues, but it has been quite aggravating.

 

---

 

 

1) is basically a "dmget" problem. dmget should be installed on discover sometime in early December. In the meantime instead of using PBS in the "acquire" script modify it to simply "scp" the data. That will the remove the 1hr wall time requirement.

 

(2) please contact NCCS directly. They should be able to increase your scratch space to whatever size you need.

We are also thinking about implementing a "scrubber" script that will get rid of "useless" files that remain after a DAS run fails.

 

 

Comments (1)

Bookmarked By (0)

More Like This

  • Retrieving data ...
USAGov logo NASA Logo - nasa.gov