I have two switches (Netgear Prosafe FS104's) and want to use these to create a LAN with cheap redundancy. So, if one switch fails, the other can still route traffic between our PCs and maintain communications with the outside world. Each switch has its own internet connection to the outside world. Each PC has only one NIC. I was thinking of splitting the 2 pairs of TxRx lines available on each NIC and sending one pair to each switch. The question is what else would I need to get this to work? All PCs run Linux (Ubuntu).
End-user computers, generally, do not require this level of redundancy. You would see this level of fault tolerance in a stacked/chassis setup for something like a server farm. You would have two routing engines, two power supplies, multiple cards and multiple links to that node with 802.1ax (i.e. LACP/PAgP) to help protect you in every instance from an outage.
If you want redundancy, split the network into two. Have half your connections going to one switch, and the other half going to the former.
I was thinking of splitting the 2 pairs of TxRx lines available on each NIC and sending one pair to each switch.
This is a very bad idea; not to mention it will not work how your anticipating it will. If you did split the pairs like this, you would only get use out of one set of them. Auto negotiation will allow a single pair to function by autoing down to 100Mb; remember, 100BASE-TX only functions on pairs 1,2,3 and 6.
I did some research, and one solution to my problem is to use Parallel Redundant Protocol (see http://wiki.wireshark.org/PRP). This requires a dual-NIC node to participate in the network, but single-NIC can too by using a Redundancy Box or "RedBox", which duplicates an incoming packet and sends it to the two parallel networks. There are commercial RedBox products (eg: http://www.ruggedcom.com/pdfs/datasheets/rs950g_datasheet.pdf), but they are unlikely to be "cheap". I'd need RedBoxes for each participating single-NIC node.
The split-network solution suggested by Fizzle is also not as bad as I thought -- it handles all single-point failure modes. Also, if the FS104's are more reliable than the datalogger & processing nodes, the split network would also handle the 2-point failure mode in my last comment. All I have to do is to write the L3 apps to handle this type of failure.