VMWare VM network kernel parameters change doesn't any performance improvement

[therealhansolo]

Hello all,

  I'm having TX drops on VM interface, when i'm sending some traffic throught that interface. So i tried to tune the kernel parameters for the interface, like setting the ring buffers, changing the read/write buffer memory to more than 2 MB which is the default. But none of the changes helps in lowering the TX drops.

Anyone have any idea how to lower the TX drops ?

Thanks,

Hanish

[SureshKumarMuth]

Hi Hanish,

You mean the Tx drops occurs at the virtual nic level of the VM ? Tx drops generally occur when the source send the data faster than the device which is receiving it. After making changes in ring buffer do you see any reduction in drops or still the same ?

Are you using E1000E or vmxnet3, have you performed network test with the VMs running on same portgroup in same vSwitch ? this will help us to isolate if the packet drops occurs only when the data goes out of physical card.

[therealhansolo]

Hello Sureshkumar​,

Yes, the driver is details is:

driver: vmxnet3

version: 1.4.14.0-k-NAPI

VM is running on the same port group and the same vSwitch.

The drops are around the same.

The drop stats are follows:

-----------------------------------

Drop is 0.2889893187955948 %

when the parameters are;

Ring parameters for ens192:

---------------------------

Pre-set maximums:

RX: 4096

RX Mini: 2048

RX Jumbo: 4096

TX: 4096

Current hardware settings:

RX: 4032

RX Mini: 128

RX Jumbo: 256

TX: 256

Socket receive buffer

---------------------

net.core.rmem_default = 212992

net.core.rmem_max = 1073741824

===============================================

Drop is 0.4940815265973368 %

when the parameters are;

Ring parameters for ens192:

---------------------------

Pre-set maximums:

RX: 4096

RX Mini: 2048

RX Jumbo: 4096

TX: 4096

Current hardware settings:

RX: 4032

RX Mini: 128

RX Jumbo: 256

TX: 4096

Socket receive buffer

---------------------

net.core.rmem_default = 212992 bytes

net.core.rmem_max = 1073741824 bytes

Thank you,

Hanish

[SureshKumarMuth]

What is the Rx ring buffer settings in the destination machine . Do you see any Rx drops there ? is the destination machine is slow in processing data ?

Check if this article help you to dig further at different levels (Guest OS, ESXi etc)

https://vswitchzero.com/2017/09/26/vmxnet3-rx-ring-buffer-exhaustion-and-packet-loss/

[therealhansolo]

Okay. Thank you. No, i cannot see any RX drops on the other machine.