I have installed Solaris 10 U6 on VMware VI3 ESX 3.5 U4. Both are fully patched and as up to date on drivers as possible. The server hardware is HP DL580 connected to the network with copper and 6 NICs or a HP Blade environment using the Flex 10 virtual connection. The core switch is a Cisco 6509. To my knowledge all firmware is either up to date or in full compliance with the VMware HCL. Solaris is using the E1000g driver and is in 64-bit mode.
The Solaris VM on very random occasions will drop SSH sessions and I can see it lose pings. It is very intermittent and the way it fails would almost look like hardware. A continuous ping can work for long periods of time then fail 5 times, continue on again and so on. This does not seem to have a specific pattern but causes a lot of performance issues and connectivity loss. I have VMotioned the VM between the DL580s and the Blades and I have the same problems. No matter where I put Solaris 10 in my VMware environment I have performance and connectivity issues. Windows and other servers in this environment are not affected.
We have tried Solaris 10 on bare metal with the same results. We have run every update Sun has to offer so we are at the most current patch and driver levels available. In doing research at the switch level we noticed that the MAC address for Solaris servers keep dropping from the CAM table. Our current workaround is to statically map the MAC in the CAM or to have the Solaris box run a continuous ping to its gateway. Obviously neither workaround is optimal.
Does anyone know of a setting on the NIC or something I might be missing? Why would only Solaris 10 servers drop their MAC from the CAM table? Any help would be appreciated.