Hi All,

I am running Fluent 6.2 on a 9 node server with Gigabit interconnection and running Linux Redhat 8.

Recently a large number of files have been giving grid errors when reading cases into Fluent. I beleive the problem is to do with Linux nfs.

When files are written as gzip files no such problems are encountered.

I am wondering has anyone else ever had similar problems?

Thanks

I'm not sure what your problem is exactly and how you easily connect nfs with files giving you grid errors. Can you explain in a little more detail?

The problem first appeared in our large case files. approx 1.5-2 million cells. The errors seem to happen at random. For example, I set up a case in serial and then in parallel and iterated the case to get soltuion. Write the case/data files. Then exit Fluent. If i start another fluent session and try and read case back in i get "grid build error" or "null domian pointer".

Also there is another reason. Some home directorys are mounted on different nodes using nfs. For one particular file when i perform a checksum of the file on the master node and compare this to checksum of same file when i ssh into node of home directory, i found different checksum. This is why i think its nfs problem.

As i said in previous message this problem dissapears when using gzip. I think this is because gzip performs checksum on files when writing them.

I am not an expert on Linux but I dont think it is a Fluent problem.

The easiest and trouble free setup in my experience has been to have machine1's working directory say /home/user exported via nfs to all the nodes. Next ensure that one can rsh/ssh between any 2 nodes without a password (This I guess, must have been arranged in your setup already). Once this is done, all writing will be done on one filesystem on machine1, a mirror image of which is available on all nodes.

Thanks for reply pUl| We have this already set up and still get problems when using the filesystem on machine1.

We are using the filesystem on other nodes due to space requirements. Going to rebuild the system with newer version of linux over next few months.

This descirbes someone with similar problem http://www.beowulf.org/archive/2002-May/007151.html

If I understand you correctly, the files themselves are being written out in parts - individually by each processor and later, after convergence their concatenation is attempted. This might sound credulous, but I do hope that you are writing 'binary' files out, not ASCII.

1)I am also facing a similar problem. My mesh size is 4 million with all modes of HT, shell conduction, grid interfaces etc. I am not able to run any of the autosaved case & data file. Version:- 6.2.16 2) sometimes not able to write or read .gz file. What is the remedy for it.

Mahesh