Re: [tcpm] some questions related to PAWS

[Joe Touch <[e-mail]>]

On 6/2/2013 7:02 AM, Yoshifumi Nishida wrote:
> Hi,
> I start wondering the following points on PAWS.
> Please let me know if I miss something.
>
> 1: Do we always need PAWS check?
>      If TCP uses enough length of MSL and it hasn't sent 2**32 data
> yet,  I personally don't see strong reason to drop the packet without
> TS-opt.
>
>     Also,  when TCP has sent more than 2**32 data and if it has good
> knowledge that more than MSL time has passed since a packet with the
> seqno has sent, do we really need to perform PAWS check?

The trouble is that we don't have a way to say "use TSopt to measure 
RTT" vs. "use TSopt for PAWS".

Right now, AFAICT it means "use TSopt for PAWS and to measure RTT".

However, sending more than 4GB isn't as absurd or exceptional as it used 
to be, and lots of devices do lots of things with TCP packets that keep 
them around a while. It'd be nice if "MSL" was relevant here, but 
without TSopt, AFAICT MSL isn't involved except voluntarily at the 
source; if a source wants to send more than 4GB in 2 minutes (a transfer 
rate over 267Mbps, which is easy even on a laptop), then it can, will, 
and *does*.

> 2: Do we always need to enable both RTTM and PAWS when TS is negotiated?
>      I'm wondering there might be cases where sender only wants RTTM or PAWS.
>     Since the sender's choice doesn't affect the receiver's logic,
> implementing a simple parameter at the sender might be enough for
> this. Or, does it cause serious problems?

I'm not sure how to do that (see above).

> 3: Action for ACK without TS-opt
>      I'm guessing we don't need to send an ACK for the last in-sequence
> segment when a packet without TS-opt is ACK. But, I think the draft is
> not very clear on this point.

It should be more clear; IMO, all segments in a TSopt-negotiated session 
without TSopt should be silently ignored, but that's me ;-)

Joe