[Sis-uce] Thoughts?

[Krupiarz, Christopher]

Scott,

I was thinking of operating serially although I hadn't thought of the
advantages you mention for a parallel timer.  I envisioned that once the EOF
is received, the Check Timer is started for a bit to get any stragglers,
then, if the file is still not complete, start the NAK process.  Thus no
NAKs are sent in the most likely case.  With a link without retransmissions,
the Check Timer would be 0.

However, unless someone can come up with a compelling example of using
acknowledged mode over a link with retransmissions, I'm all for minimizing
complexity and leaving it out.  Even then, the worst case scenario is
probably the transmission of one NAK since I don't see the NAK timer set so
tight that it didn't get the rest of the PDUs that were still in
transmission before the NAK timer timed out (this skips over why the NAK
would have even timed out to begin with since delivery is guaranteed by the
link layer--this stuff can really make your head spin :).

Chris

-----Original Message-----
From: Scott Burleigh [mailto:Scott.Burleigh@jpl.nasa.gov]
Sent: Thursday, May 22, 2003 12:20 PM
To: sis-uce@mailman.ccsds.org
Subject: RE: [Sis-uce] Thoughts?


At 11:28 AM 5/22/2003 -0400, Krupiarz, Christopher wrote:
>All,
>
>The following is an exchange regarding Scott's concept that he suggested be
>forwarded to the mailing list.  I'll put my NAK comments here so that it
>doesn't get too garbled:
>
>The NAK timer would serve pretty much the same thing, but it would send a
>NAK that may be unnecessary in the event of late arriving PDUs.  Thus the
>sender would end up using bandwidth to resend PDUs that the receiver was
>probably shortly to receive.  Of course, this is all dependent upon someone
>using acknowledged mode which, as you say, isn't likely.  Therefore it's
>probably not worth introducing the added complexity to acknowledged mode.

Okay, I think I see your point: if there were a Check cycle timer operating 
in parallel with the NAK cycle timer in acknowledged mode, but on a much 
shorter period, then it could detect file completion sooner without 
increasing the risk of issuing an unnecessary NAK and triggering 
unnecessary retransmission.  It would reduce NAK traffic if you would 
otherwise run with a NAK timer period so short that premature re-NAKing was 
likely (it would let you increase the NAK timer period without increasing 
delay in delivering completed files); if you were already using a 
long-period NAK timer, then it would reduce file delivery latency.

There's certainly value in this; the question would be whether that value 
would justify the additional complexity in the spec and in the 
implementations.  I'm inclined toward minimizing additional complexity in 
CFDP at this point, even at the cost of sub-optimal performance, but what 
do you guys think?

Scott

> > -----Original Message-----
> > From: Scott Burleigh [mailto:Scott.Burleigh@jpl.nasa.gov]
> > Sent: Thursday, May 22, 2003 10:57 AM
> > To: Krupiarz, Christopher
> > Subject: RE: [Sis-uce] Thoughts?
> >
> >
> > At 09:16 AM 5/22/2003 -0400, you wrote:
> > >Scott,
> > >
> > >FYI, I see no problem with the concept.  It sounds like it's
> > needed and this
> > >looks like a good solution.  One question: do you see any
> > scenarios where
> > >acknowledged mode would still be used even with a
> > retransmitting link layer?
> >
> > Actually, no.  If the link layer is reliable, I think
> > acknowledged-mode
> > CFDP would be unnecessary overhead.
> >
> > >I was trying to come up with some but none seemed terribly
> > realistic.  I was
> > >thinking along the lines of maybe a sensor web or a rover
> > that couldn't
> > >handle the extra complexity may communicate to some waypoint
> > that was using a
> > >retransmitting link layer.
> >
> > I suspect that the link-layer retransmission being proposed
> > at JPL will be
> > simpler than acknowledged-mode CFDP, but I agree that the
> > scenario you're
> > talking about is possible.  If the rover is using ack-mode
> > CFDP over a
> > non-retransmitted link, then the waypoint will have to be
> > doing the same
> > thing in order for the two to communicate.  The question then
> > is what the
> > waypoint-to-earth communication looks like.
> >
> > If you're using Extended Procedures then you've got only a
> > single CFDP
> > transaction end-to-end; I'm pretty sure you'd have to stick with
> > acknowledged mode on the waypoint-to-earth link even if that
> > link layer had
> > its own retransmission.
> >
> > I think that's in fact an argument against Extended Procedures that I
> > hadn't considered.  If you are instead using
> > Store-and-Forward Overlay,
> > then the waypoint-to-earth transmission is a separate
> > transaction with a
> > potentially different configuration: it could be
> > unacknowledged-mode CFDP
> > over the retransmitting link.  A little cleaner.
> >
> > >The reason I ask is that if there were, it would
> > >naturally be beneficial to have the Check Timer on
> > acknowledged EOFs in
> > >order to reduce NAK traffic.
> >
> > Can you tell me a little more about how you'd see this
> > operating?  I'm
> > thinking that in acknowledged mode we already have the NAK
> > timer doing
> > essentially the same thing (i.e., it starts when you receive
> > an EOF and not
> > all of the file data and metadata have been received).
> >
>
>
> > Scott
> >
> > P.S.: please don't feel inhibited about posting this kind of
> > exchange to
> > the sis-uce mailing list.  This stuff is highly relevant to
> > the charter of
> > the BOF and the future WG.
> >
>_______________________________________________
>Sis-uce mailing list
>Sis-uce@mailman.ccsds.org
>http://mailman.ccsds.org/mailman/listinfo/sis-uce

_______________________________________________
Sis-uce mailing list
Sis-uce@mailman.ccsds.org
http://mailman.ccsds.org/mailman/listinfo/sis-uce