Here I describe our experience in setting up a cluster for scientific
computing.
It's not exactly addressing the problem you describe, but Rendezvous
works fine on my private network, even from the Head Node which is on
the WAN side and on the private network.
TG
We have set up an Xserve Head Node with 8 Xserve compute nodes. The
Head node connects to the other Compute nodes in the compute cluster
via the 2nd ethernet port (en1) and a fanned ethernet switch - this
forms a local 10.0.0.x subnet.
The 1st ethernet port (en0) of the Head node is used to connect to
the outside world.
The XGrid Controller is being run on the Head node. The Rendezvous
logic in XGrid does not appear to be finding the other compute nodes
on the local 10.0.0.x subnet. I assume this is because the Head Node
is only looking on the en0 port to the network outside, and not on
the en1 port to the cluster subnet. Is this something that can be
patched easily in XGrid?
If the XGrid Controller is run on a compute node which only sees the
local subnet 10.0.0.x then the XGrid auto discovery works. However,
this compute node is not directly visible to the outside world, so
submitting jobs is more involved.
Yours, Rick Gaitskell
_______________________________________________
xgrid-users mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/xgrid-users
Do not post admin requests to the list. They will be ignored.
--
_________________
Terry M. Gray, Ph.D., Computer Support Scientist
Chemistry Department, Colorado State University
Fort Collins, Colorado 80523
email@hidden http://www.chm.colostate.edu/~grayt/
phone: 970-491-7003 fax: 970-491-1801
_______________________________________________
xgrid-users mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/xgrid-users
Do not post admin requests to the list. They will be ignored.
