While we used Nova as computing nodes on OpenStack, the network will malfunction occasionally. The weird thing is the configuration of Ethernet (i.e., eth0) still correct. However, we yet figure out the true cause. The alternative solution is to check the functionality of Ethernet (e.g., ping somewhere) periodically. For example, I put the following command into
/etc/crontab in CentOS 6.x and check the network per minute; If encounter the failure, system will restart the network.
*/1 * * * * root ping 220.127.116.11 -c5 -w5 || /etc/init.d/network restart