10.7.5 host, 10.8.5 guest. Fusion 6.0.2
host details: xserve2,1 8 cpu cores, 4GB ram. Ethernet: AppleIntel8254X ver 2.1.3b1 (2 cards)
bridged networking. When copying large disk images to AFP share to guest or backing up guest with netbackup (I think the commonality is that there is a single continuous stream of data from a large file), the network connection fails after some amount of data, often around 40GB but sometimes more or less. The problem goes away with NAT networking using forwarded ports via the nat.conf file for vmnet8
The symptoms are, at first seeing a dozen or so of these kernel messages on the host:
vmnet: Failed to deep copy mbuf: 12.
vmnet: bridge-en0: SendToVNet copy failed: 12.
Then a few minutes later the network copy/stream just stops. Other network traffic to the guest is fine (ssh sessions are not interrupted, etc). Netbackup logs indicate a network connection timeout. On the osx client copying the data to guest via afp, the finder hangs and is almost impossible to abort or cancel the upload.
After tearing my hair out trying different settings, nic options, kernel options, and even limiting the bandwidth to 10mbit via the vmx file ethernet options, and still having the problem-- I changed the vmx adapter from e1000e to just "e1000". The performance seems to be about the same and I still get kernel errors on the host, but the network streams remain flowing. In fact, I get some additional kernel errors now:
vmnet: Failed to deep copy mbuf: 12.
vmnet: bridge-en0: SendToVNet copy failed: 12.
vmnet: Failed to deep copy mbuf: 12.
vmnet: mbuf_dup() failed: 12.
There are only perhaps 50 lines of kernel messages and then it is quiet, and the network streams continue to function for hours without any additional kernel messages. I am getting a sustained send speed of 60-65 MB/s from my guest's perspective during the net backup operation.