Re: [trill] Fwd: Mail regarding draft-ietf-trill-over-ip

["Templin, Fred L" <[e-mail]>]

HI,

> -----Original Message-----
> From: Xuxiaohu [mailto:[e-mail]]
> Sent: Tuesday, May 05, 2015 7:47 PM
> To: Joe Touch; Tom Herbert; Templin, Fred L
> Cc: Donald Eastlake; [e-mail]; [e-mail]; [e-mail]; [e-mail]
> Subject: RE: [trill] Fwd: Mail regarding draft-ietf-trill-over-ip
> 
> 
> 
> > -----Original Message-----
> > From: Joe Touch [mailto:[e-mail]]
> > Sent: Wednesday, May 06, 2015 3:42 AM
> > To: Tom Herbert; Templin, Fred L
> > Cc: Xuxiaohu; Donald Eastlake; [e-mail]; [e-mail]; [e-mail];
> > [e-mail]
> > Subject: Re: [trill] Fwd: Mail regarding draft-ietf-trill-over-ip
> >
> >
> >
> > On 5/5/2015 12:34 PM, Joe Touch wrote:
> > >> Or just define a simple version translation as part of encapsulation.
> > >> > So for IPv8:
> > >> >
> > >> > 0x1000->0x0101 on encapsulation
> > >> > 0x0101->0x1000 on decapsualtion
> > > And what happens to 0x0101 WHEN it shows up?
> > >
> > > You need more patterns than you have because IP is allowed to use any
> > > of them.
> >
> > FWIW, you're essentially arguing for bit-stuffing, i.e., if a pattern you think will
> > be common shows up, don't stuff, and if something else shows up, then do.
> >
> > That works only if you do true stuffing, e.g.,:
> >
> > 	0x01** = uncompressed
> > (might as well add 0x11** and 0x10** to that)
> > 	0x00** = compressed
> >
> > But then if 0x0000, 0x0001, 0x0010 or 0x0011 show up, you need to decide
> > how to represent them - they CANNOT be uncompressed without at least two
> > more bits somewhere else that indicates they're uncompressed and their value.
> >
> > So now your GUE methods are very sensitive to IP version numbers, which IMO
> > defeats the "G" in the name. Never mind how this complicates hardware when
> 
> Agree. As I had mentioned to Tom in private, I do believe the additional capability of GUE (e.g., fragmentation) would be much useful
> in some network environments. However, since GUE is intended to be a generic UDP encapsulation approach, we'd better design it as
> generic and future-proof as possible from the beginning, specifically speaking, without being deeply influenced by a particular GUE
> payload type. If both foo-in-GUE (foo could be NSH, TRILL...) and foo-in-UDP (i.e., a compact version of foo-in-GUE) are needed
> someday, could you still achieve the same effect as that for IP-in-UDP (as a compact IP-in-GUE) and IP-in-GUE?

I don't get why we are still talking about this. We have already shown how
IP-in-UDP can be properly accommodated by GUE with header compression.

Thanks - Fred
[e-mail]

> Best regards,
> Xiaohu
> 
> > (not if) other IP versions are used (for whatever reason).
> >
> > Joe
> >
> >