Hi Dino,

I agree with both of your points, my concern was that the moment the data plane 
traffic hits user space we degrade performance. Not a problem though if we're 
just prototyping __

I'm trying to think of a solution that does not need to modify WG but I can't 
come up with any.  For example if we want to control the src UDP port that WG 
puts in the packets, we need a way to specify this, and I'm not aware you can 
do it (from userspace).

Thanks,

Jordi

On 3/24/20, 14:32, "Dino Farinacci" <[email protected]> wrote:

    > Yes, we considered adding an IID. However, since this means changing the 
WG kernel code, we discarded this option in favor of a user-space solution. I 
agree that it would be a nice addition though :)
    
    How about use source-port in UDP header. That can be done in user space. 
You lose load-balancing but you caould use 10-bits for IID and 6 bits for 
entropy.
    
    > From an implementation perspective I don't think it's straightforward to 
do LISP in WG encapsulation without substantial processing overhead. Right now 
we're programming the WG interface, but if we want LISP in WG we probably need 
handling packets in user space also.
    
    Well I would think the overhead would be the same as any other 
encapsulation. But to compare apples with apples, you’d have to put the LISP 
encapsulation in the kernel as well. So back to your first point.
    
    Are there any spare bits in the WG header. Which is the default 
encapsulation it uses. If its IPsec, there are plenty of usable bits in the 
ESP/AH header. ;-)
    
    Dino
    
    


_______________________________________________
lisp mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lisp

Reply via email to