CRIO disconnects from Ethernet network and stops responding to pings

Question

We are running a cRIO application that periodically disconnects from the Ethernet network even though the VI continues to run on the cRIO.
The application has two parallel loops.  One loop reads data from the FPGA and writes it to a RT FIFO.  The second loop opens a TCP connection to a remote host, reads data from the RT FIFO, and transmits the data to the remote host.  We have instrumented each loop to send messages to the serial port that indicate their status.  When the cRIO disconnects, both loops continue to send their status messages to the serial port. The first loop happily continues reading data from the FPGA and writing it into the RT FIFO.  The RT FIFO overwrites the oldest data when it fills up to keep it from overflowing.  The second loop reports a TCP write timeout (tcp write error 56) around the time that the cRIO disconnects, after which it closes the existing TCP connection and then repeatedly tries to open a new TCP connection (each attempt times out after a 5 second delay).
All that appears normal and usually the connection is automatically restored.  Occasionally, however, the cRIO is not able to re-open the TCP connection.  Furthermore, the cRIO stops responding to pings from the remote host.  On two occasions, the cRIO was finally able to open a new connection after 30-40 minutes.  Other than that, the only way we've found out of this dilemma is to reboot the cRIO.
The frequency of the disconnects and the time before they occur doesn't seem connected to whether or not the cRIO is set to run the application at startup or if the application is run via an interactive VI window on the remote computer.
It is possible that the disconnects have something to do with which computer the cRIO is connected to, as the problems seem to happen much more frequently (often within minutes) when the cRIO is sending data to a laptop running Windows XP, but it can run for hours when sending data to a rackmount server running Windows Server 2003.  Everything else (the speed and size of data being sent, the ethernet hub and cables, the VI running on the remote computer) is the same.

Answer 1

We're done most of the testing with a 9014 real time controller and a 9104 chassis.  However, we've also seen similar behavior with 9014/9103 and 9102/9101 combinations.
The 9014 cRIO normally runs at around 50% CPU usage and less than 40% memory usage when the data is flowing steadily.  The RT System Monitor did not indicate a spike in either before the cRIO disconnected.
Measurement and Automation Explorer is not able to connect to the cRIO after it disconnects.  We tried pinging the cRIO from a different computer, but that also failed.
After disconnecting, the cRIO sends error messages to the serial port while trying to open a new TCP connection. It reports an error 56 (time-out error) 3 or 4 times (for a total duration of 15 or 20 seconds) and then reports an error 42 (generic error) at least 10 times per second for 18 seconds, and then switches back to 15 seconds of timeouts, and so on. When getting error 42, it ALWAYS gets it for 18 seconds straight.  It usually gets error 56 for 15 seconds (3 timeouts) but sometimes for 20 seconds (4 timeouts).
When the cRIO disconnects, it can't be contacted by any computer through the ethernet port, yet the link lights still show that it is connected. If its connected directly, both the green and orange link lights on the cRIO light up, but when it is connected through the 10 Mb hub, only the orange link light lights up, but the hub shows that the cRIO is connected to it. In spite of this, it is still impossible to ping the cRIO in either scenario.
It is possible that, in addition to what computer the cRIO is connected to, the errors could be related to whether the cRIO has a direct connection to a computer or if it is connected through a hub. Today, we have only managed to get errors when the cRIO is directly connected to a laptop (one error after 1.5 minutes, one after 10 minutes, one after 7 minutes, etc.)  There were no disconnects when the cRIO was connected to a 10 Mbps hub for more than half an hour.
Do any of those symptoms give you clues about what is happening?
Thanks!