[Beowulf] motherboards for diskless nodes

Tony Travis ajt at rri.sari.ac.uk
Sat Feb 26 03:51:34 PST 2005


Ryan Sweet wrote:
> [...]
> I think the term "diskless" is sometimes the problem when discussing
>  centrally installed and managed systems.  Lots of "diskless" cluster
>  have GB and GB of local disks, only they are used for swap and temp
> I/O, not for the OS.

Hello, Ryan.

I agree with you: It is common for so-called 'diskless' nodes to have a 
local disk for /tmp and swap. In our case I have also made a symbolic 
link /var/tmp -> /tmp on each node as well. In fact, Sun used to call 
this a 'dataless' client (i.e. no permanent data stored on the client, 
only local swap and temporary files). In the end, Sun abandoned support 
for 'dataless' clients in favour of their NFS-based cacheFS.

The important thing about diskless/dataless/cacheFS clients is that they 
can easily be replaced with a new one if they go wrong without loss of 
permanent 'data'. Of course, the data associated with processes actually 
running on a node is lost if the node fails in use, but the new node is 
a plug-in replacement for the old one, and just needs to be rebooted. In 
our case, there is a little bit more to it than that because we have to 
add the new MAC address to the DHCP server for the fixed IP address used 
by the node, and partition/format the new local disk but this is done by 
a script and takes about 2 minutes!.

It might be a lot less confusing if we talked about PXE booting dataless 
clients/nodes...

	Tony.
-- 
Dr. A.J.Travis,                     |  mailto:ajt at rri.sari.ac.uk
Rowett Research Institute,          |    http://www.rri.sari.ac.uk/~ajt
Greenburn Road, Bucksburn,          |   phone:+44 (0)1224 712751
Aberdeen AB21 9SB, Scotland, UK.    |     fax:+44 (0)1224 716687



More information about the Beowulf mailing list