First of all please verify whether your tx buffer has the same size as the rx buffer.
If the tx buffer is bigger than rx buffer, then your VMs might not be able to keep up with the transmit.
I saw this couple of times, especially after migrating to the environment with high speed uplinks.
Second thing is having proper vmxnet3 drivers.
Do you have it without vm tools ?
Are there open vm tools for free bsd ?
E1000 is a good workaround.
This driver has really big symmetrical buffers and should be available by default in the OS.
Thanks for the response.It might be right because before VM migration to vSphere 6.7,it was running(VM network) with 2 * 1 G physical nics and now its 2 * 10 G nics(VM network).I mean in vsphere 6.7
As said earlier, we have not installed any vmtools and opentools on freebsd since beginning.
Do we need to install specific vmxnet3 drivers? When we install opentools, vmxnet3 driver will automatically come along this?
I will check the tx and rx size and update you.
Open VM Tools will contain vmxnet3 driver among the other things that are nice to have on the VM.
Why are you so eager to run vmxnet3 with a FreeBSD guest ?
I would suggest to switch to a nic that comes with stable FreeBSD drivers - if available for FreeBSD I would try the e1000e first (thats also a PCIe device)
Thanks for the response.
We installed openvmtools on two vms(vmxnet3) but no luck.
We downgraded from vmxnet3 to e1000 but no luck.
We have not adjusted the TX and rx values so far.
Now we migrated other set of windows(vmtools up to date) vms to new 6.7 infrastructure where we are seeing the same issue.
We have 4 to 5 vm vlans.
Vlan id 100 to 105
When we ping the same vlan subnet vms (ex 100), there is no issue. Freebsd vm1 has 10.10.100.5 and windows vm has 10.10.100.6 ---- No packet loss.
When we ping the different subnet vms(vlan 100 to 101,etc....), we are seeing packet loss. I mean 5 or 10 percentage packet loss.
This is cross vlan issue and don't know what is the problem behind.
We migrated some vms back to old vSphere 5.5 infra where there is no cross vlan issue and no packet loss reported among different vlans.
We are using ip hash load balancing with static ether channel.
So what is the router between the VLANs?
It's a Cisco router.
I got some clue. From this first article it seems this is known issue in vsphere 6.7 update 1 express patch 7
From this below community, it seems this is known issue in vsphere 6.7 update 1 express patch 5.
My doubt:- I'm using vsphere 6.7 update 1.Is this known issue in the update 1 as well? Any ideas?
The below link and article relates to vsphere 6.7 update 2 release note. As per this below article, it has been fixed in vsphere 6.7 update 2.
PR 2238134: When you use an IP hash or source MAC hash teaming policy, the packets might drop or go through the wrong uplink
When you use an IP hash or source MAC hash teaming policy, some packets from the packet list might use a different uplink. As a result, some of them might drop or not be sent out through the uplink determined by the teaming policy.
This issue is resolved in this release.