Sockets Applications Do Not Switch to Second NIC if One

• Microsoft Windows NT operating system version 3.1
• Microsoft Windows NT Advanced Server version 3.1
• Microsoft Windows NT Workstation version 3.5
• Microsoft Windows NT Server version 3.5

SUMMARY

If a computer running Windows NT has multiple network adapters, each connected to the same physical subnet through additional hardware (such as a switch), sockets applications use only one of the adapters to communicate over the network. In addition, if the intermediate hardware fails, sockets applications do not detect the failure and do not attempt to recover by using another adapter/switch combination.

MORE INFORMATION

You can use network hardware in parallel configurations to provide fault tolerance. For example, you can put two network adapters in the same computer, connected to the same physical subnet, so that if one adapter fails the other can take over. If you need intermediate hardware, such as a switch, to connect each adapter to the network, you can attach one switch to each adapter, rather than connecting all the adapters to the same switch.

This strategy assumes that the applications communicating over the network can detect the failure of one set of network hardware and automatically start using the other one. Unfortunately, many sockets applications do not detect hardware failure, and will not automatically use the other set of hardware. For example, you run the LPR print monitor on a Windows NT print server with two network adapters, each with its own switch. When LPR initializes, it makes sockets calls that complete correctly as long as one of the network adapters is functioning, even if the switch connecting that adapter to the network has failed. LPR has no way to determine if the failure is with the switch, rather than with other hardware on the network. Therefore, LPR has no reason to try the other adapter/switch combination. While LPR is running it can tell when it loses connections with LPD servers, but it does not know why. If also doesn't know to try using another adapter to work around the problem.

NOTE: This problem applies to hardware connecting the computer to the network, not to other devices on the network, such as routers. For example, if there are multiple routes from the Windows NT LPR client to an LPD server, and one route fails, TCP/IP will automatically use a different route, and LPR will continue sending its print jobs.