semi-solved now.

Error disappears after:

1. I deleted all other mpich variants from all .bashrc (on second node there
was intel mpich one)
2. I deleted all "localhost" strings from machinefile and copied machinefile
to all nodes (not sure is it necessary)

After that I cannot say that NFS always slows performance.

For ~/machinefile

node1:8
node2:8

Calculation started from node1 gives:

With NFS sharing of siesta and mpich dirs: 231
Without NFS sharing of siesta and mpich dirs: 229

But for ~/machinefile

node1:0
node2:8

Calculation started from node1 gives:

With NFS sharing of siesta and mpich dirs:  3412 (sic!)
Without NFS sharing of siesta and mpich dirs: 239

So it seems to be not NFS only stuff but complicated
node-device-process_manager-load_balancing stuff.



Mike Turchenkov
PhD Student, MEPHI, Moscow.

Responder a