@gestinfo
Don't know if this will help you or anyone else with ESXi and an ISCSI initiated at the VM level.
Since I added custom TCP/IP stacks and a VMKernel to our vSwitch for the backup vlan, stability under healthcheck loads is a lot better.
I suspect, (any Network Hardware boys and girls out there feel free to comment) that the reason is two fold.
Under normal configuration the default TCP/IP stack is on the Management IP VLAN and therefore all other internal VM traffic gets a default route on the Management IP. Externally I suspect this meant the Switches which are layer 3 aware was being forced to bridge VLans and route between the Management LAN and Backup LAN for backup traffic. During healtcheck we would get Managment warnings for disconnected hosts as it saturated bandwidth.
The second reason I think it helps is something that Gostev confirmed after I questioned the comment of one of the support engineers as to why my network performance on throughput was way below theoretical maximums. ESXi has a bug in that adding a VMKernel to a vSwitch limits throughput to 45% of physical capacity inside ESXi.
So adding the VMK at the vSwitch level on my Backup LAN effectively throttles network throughput at the Hypervisor level. I have therefore also throttled my backup network in Veeam to a little below the 45% to hopefully acheive some headroom for control and reporting during backups.
Don't know if this will help you or anyone else with ESXi and an ISCSI initiated at the VM level.
Since I added custom TCP/IP stacks and a VMKernel to our vSwitch for the backup vlan, stability under healthcheck loads is a lot better.
I suspect, (any Network Hardware boys and girls out there feel free to comment) that the reason is two fold.
Under normal configuration the default TCP/IP stack is on the Management IP VLAN and therefore all other internal VM traffic gets a default route on the Management IP. Externally I suspect this meant the Switches which are layer 3 aware was being forced to bridge VLans and route between the Management LAN and Backup LAN for backup traffic. During healtcheck we would get Managment warnings for disconnected hosts as it saturated bandwidth.
The second reason I think it helps is something that Gostev confirmed after I questioned the comment of one of the support engineers as to why my network performance on throughput was way below theoretical maximums. ESXi has a bug in that adding a VMKernel to a vSwitch limits throughput to 45% of physical capacity inside ESXi.
So adding the VMK at the vSwitch level on my Backup LAN effectively throttles network throughput at the Hypervisor level. I have therefore also throttled my backup network in Veeam to a little below the 45% to hopefully acheive some headroom for control and reporting during backups.
Statistics: Posted by NickKulkarni — Jan 29, 2024 1:44 pm
