Hello,
In many instances, I have noticed that parallel runs sometimes crash with the typical error message below (here partly in french, but similar pb in english):
op_read error on left context: Error = -1
unable to read the cmd header on the left context, Error = -1
Error posting readv, Une connexion existante a dû être fermée par l'hôte distant.(10054)
It seems that the trouble is mostly due to an unexpected loss of internet/network connection during the computation. As the mpiexec.exe requires user credentials everytime it is launched (or at least, to be registered once for all), it is not surprising this to occur when the TELEMAC case is launched from a user session belonging to some local network (company, university...). However, it is very annoying when one is expecting very long computations to achieve.
Is this correct and shared by other users at all?
Could anyone bring some more light about it and the mpiexec dependency to internet connection in particular?
There might be some aspects that are not clear in my head.
Thank you very much for your help,
Best regards,
JBS.