Dear All,
I have a small cluster running Torque, Maui and GPFS. The Torque/Maui
headnode is on the public network. The Worker Nodes are on a private
network. The CE is on a third machine (i.e. it is not on the
Torque/Maui headnode). The CE is also setup to do NAT/Masquerading
(using iptables) for the Worker Nodes. NAT would appear to work, that
is, you can ping (scp, sftp, globus-url-copy, etc) to the outside world
from the worker nodes. I've also tried this using SNAT instead of
Masequerading, and that also seems to work.
I *can* copy files *to* the SE (DPM) from the worker nodes
srmcp file:////home/isxjw/file.txt
srm://lcgse01.phy.bris.ac.uk:8443/srm/managerv1?SFN=/dpm/phy.bris.ac.uk/home/dteam/file.txt
But *cannot* copy files *from* the server to the workers:
srmcp
srm://lcgse01.phy.bris.ac.uk:8443/srm/managerv1?SFN=/dpm/phy.bris.ac.uk/home/dteam/file.txt
file:////home/isxjw/file2.txt
In both cases the srmcp command is being run on the worker node.
Wireshark shows that's there is a lot of traffic between SE and worker
nodes. So some traffic is definitely flowing. Possibly a new
connection is being made from the server to the worker node at some
point (and this breaks the NAT)? Is anyone else using NAT for worker
nodes? And if so have you seen this problem when using srmcp? Any ideas?
thanks
Jon
--
Dr Jon Wakelin
University of Bristol
H.H. Wills Physics Laboratory
Tyndall Avenue
Bristol
BS8 1TL
Tel: +44 117 928 8769
Fax: +44 117 925 5624
|