Re: [6lowapp] Where does TCP not work

[Kris Pister <[e-mail]>]

Completely agree that TCP doesn't make any sense if you're sending one 
packet per mote per day.
Maybe I misread the intent, or I'm thinking about different apps.
I'm not saying that we should use TCP for everything - quite the 
contrary, I agree with Pascal's reasons why we didn't use it in isa100.
Having said that, I am always surprised by the strong negative reactions 
that arise when TCP is mentioned in a wireless context.  I realize that 
there have been many good and painful lessons learned in the past, but 
it also seems quite clear that TCP can run successfully on low power 
multi-hop 15.4 meshes as well.  Many/most applications won't need this, 
but for at least a few it will be perfect.  Yes?

ksjp

Richard Kelsey wrote:
>    Date: Tue, 03 Nov 2009 02:32:25 -0800
>    From: Kris Pister <[e-mail]>
>
>    My opinion is that almost by definition (the "6" in the name), that TCP 
>    will be an important part of that future.  I'm 100% sure that we'll find 
>    other ways of doing things as well, but it makes me nervous when it 
>    looks like we're trying to reinvent every protocol from bottom to top, 
>    and still call it IP.
>
> Kris,
>
> I hope we don't end up reinventing anything.  If we get the
> layering right we can add a new protocols as needed and
> still have everyone agree that the result is IP.  In
> particular if 6LoWPAN and ROLL get their parts right, then
> TCP will work out to the nodes and will be used when it is
> appropriate.
>
> I think that this thread is somewhat missing the point.  The
> question isn't just whether TCP will fail to get data
> delivered in this or that situation, but also, in the larger
> picture, whether using TCP out to the nodes is a
> cost-effective way to get the job done.  If someone has 10k
> water meters sending one short packet a day, they may not
> want to invest in the infrastructure needed to keep 10k TCP
> connections open.  At some point in the process their data
> will be available via TCP, but that doesn't require an
> end-to-end TCP connnection down to each meter.  The great
> thing about IP is that you get enormous flexibility in how
> to structure networks and applications.
>
> We cannot create a whole new protocol stack with a gateway
> to convert between IP and our new stack.  What we can do is
> provide alternatives where appropriate and take advantage of
> the layering to allow people to choose which protocols they
> use and, most importantly, when and where they want to do
> any conversions.
>                                 -Richard Kelsey
>