Re: [Xen-users] Understanding netback/netfront GRO/GSO handling

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Fri, 2014-07-25 at 11:34 +0200, Kristoffer Egefelt wrote:
> Hi,
> 
> Is netback/netfront able to send and recieve offloaded packets (64k) via the 
> physical NICs ?

netfront/back is certainly capable of transferring packets across the PV
link without segmentation, and does so by default I believe, that
feature has been there since time immemorial. (IIRC it arrived either
just before or early in the old 2.6.18-xen fork days -- so mid 200x's).

What happens to that SKB once it gets injected into the dom0 network
stack is a function of the other components on the path to the outside
world up to and including the specific hardware in the NIC and the
relevant driver(s) features, but including any
bridging/forwarding/tunnelling infrastructure along the way.

> Specifically Iâm experiencing problems when packets egresses a OVS GRE tunnel 
> - then the packets are segmented to 1500 bytes.
> Traffic between VMs on the same dom0 maintains the 64k packet size.
> 
> As Iâm not sure if this is supposed to just work Iâm having trouble 
> identifying if the cause of this could be kernel, NIC drivers, netback, 
> openvswitch or GRE ...
> Iâm using Debian, kernel 3.13, OVS 2.1, Xen 4.4 and Intel ixgbe drivers.

Out of those I would be inclined to investigate GRE first, followed by
openvswitch.

Removing GRE from your tests would be the logical first step I think.

Ian.


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users