0

I've been given 34 servers (HP Proliant) to use for HPC. Due to the server owners' request, 33 of these will need to act as diskless thick clients, and one will act as DHCP and TFTP server to enable network booting for the others.

The "head" server is already configured with Debian Squeeze and is operational.

The problem lies in how the network is organized:

  • Each server has 2 network interfaces
  • Each server has only 1 IP allocated (unmodifiable)
  • Everything is under a VLAN, so all interfaces interfaces must be given the proper VLAN tag

While setting the interfaces is not difficult, the issue is how to handle the DHCP server: I'm not sure we can use the other network card, and as I said, we cannot request more IPs.

What would be a suggested course of action in this case?

John Gardeniers
  • 27,458
  • 12
  • 55
  • 109
Einar
  • 225
  • 2
  • 11
  • I don't see your problem. – Lucas Kauffman Jan 30 '13 at 08:47
  • You have a 34 node cluster and can't even use a separate switch (or one that allows port based VLAN mapping) to properly manage this thing? – Sven Jan 30 '13 at 10:20
  • I have absolutely no control over the network infrastructure, unfortunately. It's kind of "use what you have, don't mess with the rest" type (request by the servers' owners). – Einar Jan 30 '13 at 13:36

1 Answers1

1

I don't see your problem, you have a dhcp server and a few thin clients. Why would you use two interfaces?

Your PXE machine gives out IP's and allows your thin clients to boot. After being booted, your thin clients will just (or at least can) use the IP they received to boot their OS from the PXE server.

Lucas Kauffman
  • 16,880
  • 9
  • 58
  • 93