Hello Joe,

Here also, sorry for the delay!

>>>> I also clarified the case of encapsulation header size that is deducted.
>>> It's worth noting throughout that MTU is defined as the size of the payload.
>> Yes but there are two notions of payloads depending on the point of view.
>> Figure 10 can be useful:
>>      inner packet payload => iD
>>      tunnel link packet's payload => iH+iD
> 
> We should never be concerned with the E2E payload, i.e., iD. All we need
> to know is the size of the entire E2E message that a link (or tunnel,
> which is a link) must transfer.
> 
> However, I see the confusion here.
> 
> As context:
>    TCP MSS describes the payload, exclusive of IP and TCP headers and
> their options (RFC6991)
> 
>    ICMPv4  frag needed and IPv6 PTB indicates "next-hop MTU" as the
> largest IP packet that the next-hop link can forward without on-path
> fragmentation (RFC1191, RFC4443)
> 
> So MTU generally refers to the *payload* that a link or path can transfer.

I'm 100% okay. My comment is just that "payload" is misleading if the context is
not perfectly defined (transport vs. network vs. link, with vs. without 
tunnels).


> --
> 
> For this document:
> 
> path MTU should indicate the IP packet that can transit a path, and it's
> the min of the hop MTUs

**largest** IP packet...

> hop MTU should indicate the one-hop version of the path MTU, i.e., the
> largest IP packet that can *transit* over a link (it does NOT include
> the  link headers)
>    these are the ones that are 64 or 1280

rather than "it does NOT include..." I'd prefer:
"this network layer definition does not consider link-layer headers"

> link MTU should be defined more clearly as the link-layer packet that
> can transit the link
>    these must transit (64 or 1280) + link headers
> 
> We should use two different terms now:
> 
> tunnel hop MTU: the largest IP packet that can transit (inside) a tunnel
> 
> tunnel link MTU: the largest (encapsulated) packet that can transit
> between ingress and egress (in the current doc, we call this the tunnel MTU)
> 
> We should add a new term:
> 
> E2E MTU: the largest IP message that can transit between a source and
> destination, which is 576/1500 at a minimum.
> 
> And yes, a new figure can help explain all of this more clearly.

Great. Those definitions will help a lot.


> --
> 
> With regard to your suggestions:
> 
> - I don't think it's useful to define link MTU as 64/1280 because it
> isn't; it's those values + link (as noted above)

That's a good reason indeed ;-)
With the updated definitions, there won't be any risk of mistake.

> - the reassembly MTU definition needs to explain that it's at the level
> being reassembled, NOT the payload of the reassembled message. at IP
> receivers, this is the E2E MTU but not the path MTU. at tunnel egress,
> this is the tunnel link MTU (i.e., the E2E MTU between ingress and egress).
> 
> Does that make more sense?

Yes, perfectly clear.

>>> I think we do need another term for the path MTU between the egress and
>>> ingress but NOT inside the tunnel. I prefer to just call that the PMTU
>>> between the ingress and egress, though.
>> Okay, this is the regular IP vision of PMTU. If we need it, we can define it.
> 
> See above; E2E MTU between ingress and egress is different than any PMTU.
> 
> Joe

Thanks for the updated definitions.
Cheers,

  Vincent

Attachment: signature.asc
Description: Message signed with OpenPGP using GPGMail

_______________________________________________
Int-area mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/int-area

Reply via email to