Layer 3 Datacenters

[prelude]

Has anyone here run a server farm that is layer 3 only, i.e. that layer it run all the way to the hosts for redundancy (i.e. running OSPF on the host machines). Does this scale? What measures are take to make it scale. I have a situation where we have many systems that have historically been numbered with many addresses from random multiple IP blocks, so I have don't have an easy way to "vlan off" hosts to reduce broadcasts and interface loads. It's an interesting situation, and renumbering would be a yearlong project at this point (hundreds of machines).

andrew

[bbaley]

Would you explain with more details of your question, The host or the system mentioned in the question was refered to the PC or the router.

[prelude]

I am looking for a more flexible way to configure a server farm that is doing web hosting. Historically the machines are on a /19 that is broken up into /24s. Over the years (before I entered the picture) the techs have added ips to each hosts from random /24s. There is one /24 address on each host that is used for outbound traffic (i.e. has the default gateway), and the random IPs are used for inbound only. So, this is a big mess basically since there are so many hosts (many hundreds). I could have them renumber completely, but before I do I want to see if there are any emerging/existing alternatives, such as running an instance of OSPF on each host and allowing the hosts to tell the router what IPs (/32) are on each machine. The other alternative is trunking on each machine, but I shudder at the possibility.

One cool thing would be that if the machines were running OSPF I could have a "completely" layer 3 architecture (spanning tree, etc would still exist for access switch redundancy/load balancing). I was wondering if anyone has done this (installed a routing protocol on a massive amount of hosts).