Setting up nodes with built in hard disk

[Carlos O'Donell]

> With those 60 nodes netbooting, what kind of network overhead are you 
> seeing used for larger applications, and the initial boot of each 
> machine?  Are you using two seperate networks, one for NFS and the other 
> for your mpi/comm traffic?
> 
> Doug
>

We have a cluster of ~50 PARISC boxes and only /var and /home
(some are diskless) are on NFS. The NFS server _should_ really be 
running GigE, but we only have bonded 2x100MBit (We have 10/100 24 Port
Baystack 450 switches in a stack -> 2.3GBit interconnect).

And I can say that if you need to reboot the systems...
be prepared to time-step the starts. We almost always
completely throttle the NFS server bandwidth 
(PIII 2x1GHz, 1.5GB RAM). 

50 x 500k/s = 25MB/s = 200MBit 

Some optimizations are required... and thusly we are back
to 'need to setup big phat fast NFS server' tasks that 
admins are so used to ;)

c.