Dear Bartek: Thanks a lot for this. It certainly looks like you caught a bug.
We will certainly fix it very soon and let you all know the patch. Yours, Jose A. -- Jose A. Torres, Ph.D. Manager of the SIESTA Software On Fri, 2010-04-09 at 09:44 +0200, Szyja, B.M. wrote: > Hello again, > > Some time ago I have post a message to the list about the problem with the > SIESTA (built on IBM Power system) running out of luns. The original post is > below. In the meantime I have contacted the system administrator, who helped > me to identify the problem. Here is his response: > > this is an error message from SIESTA itself. You > can find it in the source file io.f after a check > of all fortran i/o units. The constants min_lun > and max_lun are set as 10 and 99 in the version > of the code that I have, which means that there > can be 89 files simultaneously open. There are > two possible solutions: > 1. increase max_lun so that you can have more files > simultaneously open. However, my guess is that your > code opens a file at every step, but never closes it > and so it runs out of units after 81 or 82 steps. The preferred > solution is therefore > 2. to look for the open files and close them at appropriate > points in the code. > > As my fortran skills suck like anything, and the SIESTA source is rather big, > I cannot do that myself. Moreover, it really seems like a bug to me, so I am > hoping that the developers would be interested in sorting this out... > > With regards, > Bartek > > > ________________________________________ > From: Szyja, B.M. [b.m.sz...@tue.nl] > Sent: Monday, March 29, 2010 11:27 AM > To: siesta-l@uam.es > Subject: [SIESTA-L] error - No luns available in io_assign > > Dear SIESTA Team, > > I am trying to run MD simulations using SIESTA on IBM power architecture. > Everything seems ok up to a moment when the jobs crash and I get the > following error message: > > No luns available in io_assign > Stopping Program from Node: 0 > > It usually happens after 70-80 succesful steps of MD. Does anybody have an > idea what might be the cause? > > Bartek
