Network Redundancy with no SPOF
Using the follwing bits of kit, you can eliminate any SPOF (single point of faliture) from your network, while still keeping multiple seperate networks (i.e have 2 different networks, one public, one private).
- 2 Good managed switches (Cisco Catalysts!)
- 2 Network cards in each server
- bonding.o and 8021q.o drivers in a recent kernel
- Linux! [1]
Once upon a time ..
There was a server called goat. goat had 2 network cards, one that lead to goats gateway so it could talk to the rest of the world, and one that went onto a mangement network so that goats owner could connect to him from a different network.
Goat was running some services that for arguments sake couldn't be load balanced/replicated/whatever. So it was of vital importance that goats network connectivity stayed up 24/7/365. Goats master was uberclever, and took network connectivity and hosting from a company that offers completly fault resileint networks. The problem was how the link between goat and the 2 ports his hosting compnay had given him stayed up all the time, every time. Even though switch reloads and crashes (hey, this is cisco!), goats master accidently unplugging the wrong cables, etc.
Take two pieces of string, and ..
Heres a network I made earlier.
-----------------
||==| Cisco C2950-24T |====> To upstream gateway
|| -----------------
|| ||
|| ||
|| || <-- eth0
|| --------
|| | SERVER |
|| --------
|| || <-- eth1
|| ||
|| ||
|| -----------------
||==| Cisco C2950-24T |====> To upstream gateway
-----------------
So, the idea is you have a trunk running between both of your switches.
blah blah, i need to finsih this :-p
Yes, you can probbaly faf around with windows lots and get it working, but, you (a) need an intel eepro card, and (b) i'd be more conceened about windows dying :)