Re: [tcpm] AccECN field order

["Scheffenegger, Richard" <[e-mail]>]

Just for clarity, the suggestion would be something like:

 Kind | Length | 1-bit ordering == 0 | 23-bit EE1B [ | 24-bit ECEB [ | 24-bit EE0B ] ]
And
 Kind | Length | 1-bit ordering == 1 | 23-bit EE0B [ | 24-bit ECEB [ | 24-bit EE1B ] ]

With the two trailing 3-octet fields being optional in both cases.

An increment in the CE counter would require at least 1+1+3+3 = 8 byte TCP option space.


Richard Scheffenegger


-----Ursprüngliche Nachricht-----
Von: Joseph Touch <[e-mail]> 
Gesendet: Mittwoch, 27. Jänner 2021 05:22
An: Vidhi Goel <vidhi_goel=[e-mail]>
Cc: tcpm IETF list <[e-mail]>; Scheffenegger, Richard <[e-mail]>; Michael Tuexen <[e-mail]>; Mirja Kuehlewind <[e-mail]>; [e-mail]
Betreff: Re: [tcpm] AccECN field order

NetApp Security WARNING: This is an external email. Do not click links or open attachments unless you recognize the sender and know the content is safe.




> On Jan 26, 2021, at 7:47 PM, Vidhi Goel <vidhi_goel=[e-mail]> wrote:
>
>> Please see my note from 12/3/20, which shows how a single codepoints would suffice without increasing length.
>
> If the TCP option code points is not a scarce resource,

They are. Very much so.

> then why not just keep it simple with "two different codepoints”? Is there a reason that you think we should use only one code point?

That’s the core of the discussion we’ve been having.

> Regarding your suggestion of using 1 bit from the 24-bit counter field, do you mean specify ordering once before the first ECN counter field or specify before every counter field?
> For example, if all three counters are present, are you suggesting, a) or b)? To me b) would make more sense but that makes the implementation complex as the fields have different length.
>
> a) Kind | Length | 1-bit ordering | 23-bit EE1B | 1-bit ordering | 23-bit ECEB | 1-bit ordering | 23-bit EE0B
> b) Kind | Length | 1-bit ordering | 23-bit EE1B | 24-bit ECEB | 24-bit EE0B

The point of the bit is to encode the different kinds. (b) does this just fine. If we want all three counters to be the same length, simply do so and require the other two “first bits” to be zero.

Joe