[PVE-User] Network optimization: ethtool and jumbo frames...

[Alwin Antreich]

Hi Marco,

I use the following in my config:

# /etc/network/interfaces
iface bond0 inet static
	slaves eth2.6 eth0.6

... #skipped some lines

        # activate jumbo frames
        post-up ip link set dev eth0 mtu 9000 && ip link set dev eth0.6 mtu 9000
        post-up ip link set dev eth2 mtu 9000 && ip link set dev eth2.6 mtu 9000
        post-up ip link set dev bond0 mtu 9000

You will see no change to the MTU on the bond and weird behavior like slow network response, packet drops, if you are
not setting it to all interfaces configured for the bond, including VLAN interfaces.

Hope this helps.

Cheers,
Alwin


On 05/16/2016 03:49 PM, Marco Gaiarin wrote:
> 
> I'm thinking about network optimization, both fiddling with ethernet
> parameters like buffers (ethtool -g) or hardware options (ethtool -k):
> someone have some clue?
> 
> 
> Also, as suggested by most docs i've enabled jumbo frames within the SAN
> and connected switches. But still the nodes have standard 1500 MTU.
> Two question:
> 
> 1) i'm using an LACP bond, and because i need to use the SAN also
>  within guests, a bridge.
>  I've to enable jumbo frame (eg, MTU=9000) within phisical interfaces
>  (ethX), the bond interfaces bondX, the bridges vmbrX or all of those?
>  I suppose all, but...
> 
> 2) there's no way, in web interface, to set 'mtu 9000' for interfaces;
>  i can add by hand, but looking at google i get some (old) hit from
>  forum and ML that make me think that the option does not work...
>  someone can provide me some info?
> 
> 
> Thanks.
>