Network speed issues

[nunhead_man]

Running a ping from our workstations, I can get lossless packet connection to the IP address of switch to which they are attached and  the NAS boxes they use on iSCSI

But the servers give me "Approximate round trip times in milli-seconds:     Minimum = 141ms, Maximum = 1394ms, Average = 831ms  - and the VMs inside them are suddenly even slower

This is both for the one on a fixed IP and the one that connects via DHCP

Each server has six RJ45 ports wired to the switch and each VM has its own port

Any thoughts what to check?

[J1mbo]

‌Are you using vmxnet3 NIC with BMware tools installed? Also update server and NIC firmware. Have a look in the server BIOS in case there is any power management settings that could be disabled, for testing.

are you using stock ESXi or perhaps some custom NIC drivers like Realtec 81xx?

[nunhead_man]

Thanks  - not sure I understand

"Are you using vmxnet3 NIC with BMware tools installed? "  Do you mean "VMware tools" ?  If so they are installed

Power management settings are disabled

And simply stock ESxi drivers on a bog standard set of Intel NIC.s  - 2 x 2 port plus one or two on the server motherboard

[nunhead_man]

Also re "vmxnet3 NIC" that is the type of virtual NIC?

If so we are using E1000 and E1000E

[J1mbo]

‌yes, the e1000 are problematic, try changing them. Also I would suggest simplifying the design, have one virtual switch, one port group, one uplink, and all VMs attached to that. Then one it is sorted add a second nic to the vSwitch for failover.

[nunhead_man]

OK - thanks  - different advice than elsewhere  in here

On one of the two boxes one virtual switch needs to be separate and E1000E as it runs a Notes server and that is problematic  with all else.

Changing the other two VM's on that box back to the default virtual switch and port group and moving to VMXNET3 seems to have little effect

[nunhead_man]

As you were - it has the effect of cutting the boxes off from  network access!

Not good advice I'm afraid

Swopping back to e1000 restores the connection to an average 794ms delay - as it was before

[J1mbo]

‌Then you need to keep looking for your problem; ping latency should be <1ms on the same segment just as for any physical host. what is the latency to the vmkernel port? What if you connect directly to the host (ditch the switch)? Simplify as much as you can then work back, and also check for driver updates for your hardware.