Agree it would run everywhere if NSH is treated as an *application*, as you 
point out.

The conflict is in how NSH wants to be transport independent. It is supposed to 
be a feature than violation of layering: proper layering vs. overhead, as shown 
below.

IP-strata | IP | UDP | NSH | 
IP-strata | IP | UDP | VXLAN-GPE | IP | UDP | NSH |

Ignoring, other encapsulation of NSH, carrying NSH directly over UDP does 
maintain that layering while removing the overhead. This draft is doing the 
right thing.

Surendra.


-----Original Message-----
From: Dino Farinacci [mailto:[email protected]] 
Sent: Thursday, November 05, 2015 4:21 PM
To: Bottorff, Paul <[email protected]>
Cc: Lucy Yong <[email protected]>; Surendra Kumar (smkumar) 
<[email protected]>; Alia Atlas <[email protected]>; [email protected]; Larry 
Kreeger (kreeger) <[email protected]>; [email protected]; <[email protected]> 
<[email protected]>
Subject: Re: [sfc] comment on draft-kumar-sfc-nsh-udp-transport

As I mentioned at the mic, if NSH runs over UDP/IP, then it can run over 
anything. And then every encapsulation spec doesn't need to special case NSH.

Like the analogy I used at the mic . why doest't VXLAN-GPE have a code point 
for DNS?  ;-)

Answer: it makes no sense. Run NSH as a transport layer client and it will work 
over everything we have already built and has a good chance to work over 
anything we will build.

NSH is no more an overlay than SMTP is for email.

Dino

> On Nov 5, 2015, at 4:10 PM, Bottorff, Paul <[email protected]> wrote:
> 
> It is definitely a useful option to run directly over Ethernet to allow for 
> small scale environments which don't need NVO3 overlays.
>  
> Cheers,
>  
> Paul
>  
> From: sfc [mailto:[email protected]] On Behalf Of Lucy yong
> Sent: Thursday, November 05, 2015 5:08 AM
> To: Surendra Kumar (smkumar); Alia Atlas
> Cc: [email protected]; Larry Kreeger (kreeger); [email protected]
> Subject: Re: [sfc] comment on draft-kumar-sfc-nsh-udp-transport
>  
> If SFC is deployed in Ethernet network, do we need NSH over Ethernet?
>  
> Lucy
>  
> From: Surendra Kumar (smkumar) [mailto:[email protected]] 
> Sent: Thursday, November 05, 2015 4:12 AM
> To: Alia Atlas; Lucy yong
> Cc: [email protected]; Larry Kreeger (kreeger); [email protected]
> Subject: RE: [sfc] comment on draft-kumar-sfc-nsh-udp-transport
>  
> I did go through the process of getting the ethertype for NSH and I also have 
> obtained a UDP port# in the past. I have to agree with Alia.
>  
> Lucy,
> I appreciate you guys taking a crack at NSH over GRE over UDP nested 
> encapsulation. It simply calls for unnecessary overhead and complexity in 
> formulating and processing such a packet along the tunnel path.
>  
> I admit i have not read your draft yet, will certainly do.
>  
> Regard,
> Surendra.
>  
>  
>  
> Sent from a thumb typed device.
> 
> 
> -------- Original message --------
> From: Alia Atlas <[email protected]> 
> Date: 2015/11/05 6:18 PM (GMT+09:00) 
> To: Lucy yong <[email protected]> 
> Cc: [email protected], "Larry Kreeger (kreeger)" <[email protected]>, 
> [email protected] 
> Subject: Re: [sfc] comment on draft-kumar-sfc-nsh-udp-transport 
> 
> <no-hats> 
> I think that getting a UDP port is a lot more straightforward than an 
> Ethertype.
> Not having extra bytes is also an advantage.
>  
> Regards,
> Alia 
> </no-hats>
>  
> On Thu, Nov 5, 2015 at 4:15 AM, Lucy yong <[email protected]> wrote:
> Hi Larry,
> 
> The benefit is to avoid working a UDP transport for NSH.
> 
> Thanks,
> Lucy
> 
> -----Original Message-----
> From: sfc [mailto:[email protected]] On Behalf Of Larry Kreeger (kreeger)
> Sent: Thursday, November 05, 2015 1:45 AM
> To: [email protected]
> Cc: [email protected]
> Subject: Re: [sfc] comment on draft-kumar-sfc-nsh-udp-transport
> 
> Hi Behcet,
> 
> I¹m not sure I¹m following what your point is.  It is true that VXLAN-GPE 
> also adds additional overhead which may not always be needed.  Carrying NSH 
> directly over UDP avoids that as well.  Lucy brought up a new option that I 
> had never heard suggested before, which was to carry NSH in GRE over UDP.  
> This adds a GRE header in between the UDP header and NSH, but in my opinion 
> doesn¹t bring any benefits - just more overhead and complication.
> 
> Thanks, Larry
> 
> On 11/5/15, 4:32 PM, "Behcet Sarikaya" <[email protected]> wrote:
> 
> >On Thu, Nov 5, 2015 at 1:10 AM, Larry Kreeger (kreeger)
> ><[email protected]> wrote:
> >> Hi Lucy,
> >>
> >> One of the motivations for carrying NSH directly on UDP is to avoid
> >>unnecessary overhead or complication.  Adding the GRE header in
> >>between does  not seem to add any additional benefit that I can see ­
> >>only additional  overhead.
> >>
> >
> >The point was not with VXLAN-GPE.
> >
> >Behcet
> >> Thanks, Larry
> >>
> >> From: sfc <[email protected]> on behalf of Lucy yong
> >> <[email protected]>
> >> Date: Wednesday, November 4, 2015 at 11:59 PM
> >> To: "[email protected]" <[email protected]>
> >> Subject: [sfc] comment on draft-kumar-sfc-nsh-udp-transport
> >>
> >> There is a gre/udp tunnel transport (draft-ietf-tsvwg-gre-in-udp-08)
> >>that  nsh can use for the transport; just need to register an
> >>Ethertype for nsh.
> >> The gre/udp transport provides all features nsh needs with additional
> >>security capability.
> >>
> >>
> >>
> >> Lucy
> >>
> >>
> >> _______________________________________________
> >> sfc mailing list
> >> [email protected]
> >> https://www.ietf.org/mailman/listinfo/sfc
> >>
> 
> _______________________________________________
> sfc mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/sfc
> 
> _______________________________________________
> sfc mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/sfc
>  
> _______________________________________________
> sfc mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/sfc

_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to