Re: [tcpm] timestamp options (was Re: New Version Notification for draft-touch-tcpm-tcp-edo-01.txt)

[Joe Touch <[e-mail]>]

On 6/2/2014 12:49 PM, Mark Allman wrote:
>
>> The TWHS establishes the TS for the new connection, which means old
>> packets will be seen as old and discarded by TS processing.
>
> OK, sure, fair enough.  But, again, we have plenty of non-timestamped
> connections seemingly running perfectly fine today.  So, it is pretty
> hard for me to get worked up about this "problem" the timestamp option
> is "solving".
>
>>>> 	- a new connection that either avoids TS in SYN or
>>>> 	uses just "use TS" (two-byte) flag can't know when
>>>> 	to engage the timestamp and what initial value to use
>>>
>>> This is just bogus.  On both counts.
>>
>> FWIW, I'm saying that receiver has these problems. You're correct
>> that the first TS value in an *acked* segment should be the one used
>> and that the sender can know when to turn it on.
>
> It doesn't matter if the receiver just follows the sender's lead.  The
> sender can trigger things when they are needed and the receiver can
> follow.
>
>> However, what if the receiver doesn't support TS at that point?
>
> If only I had said something about that in the note you are responding
> to ... oh wait, I did! :-)

You talk about limiting the rate; that doesn't help a connection that 
cannot proceed because packets with the TS option are being dropped, 
e.g., by a midbox. That's why I don't like new stuff that starts out of 
the blue mid-connection - you need to negotiate the capability in the 
SYN at a minimum.

Further, what happens to the packets? When and how do you know when the 
TS isn't working? When do you give up?

Your rate limit suggestion answers just one dimension of the problem; 
there are a lot of others that need to be addressed.

>> One question is - how many connections use TS that don't expect to
>> use long connections or large numbers of connections where the
>> sequence number wrap is an issue?
>
> I would say nearly all of them.
>
> I'd bet a beer that you go look in your favorite pile of traces and
> it'll be a Damn Long Time before you find even a potential problem that
> timestamps will solve.

1323bis talks about a list of mechanisms that rely on timestamps, more 
than just PAWS. What happens to those when TS is optional or started late?

Joe