Re: [manet-dlep-rg] TCP clients, servers, and discovery (WAS: Re: notes DLEP meeting @ IETF88)

[Teco Boot <[e-mail]>]

Op 14 nov. 2013, om 18:53 heeft Stan Ratliff (sratliff) <[e-mail]> het volgende geschreven:

> I thought about not sending this email… perhaps I should have abided that urge, but what-the-heck… ;-) 
> 
> 
> On Nov 14, 2013, at 11:12 AM, Teco Boot <[e-mail]> wrote:
> 
>> 
>> Op 14 nov. 2013, om 16:20 heeft Henning Rogge <[e-mail]> het volgende geschreven:
>> 
>>> I still prefer the UDP peer discovery sent by the radio.
>> 
>> +1
>> Followed by a TCP SYN response sent by router.
>> This enables set up of DLEP sessions without the UDP ignition packet. Routers can send the TCP SYN to the modem if they have knowledge on the modem IP address.
>> 
>> Teco
>> 
>> 
>>> 
>>> Henning Rogge
>>> 
> 
> If this winds up being the path taken, it means that modems are going to wind up juggling UDP and TCP sockets, including a TCP listen() socket.

No. Sending the UDP multicasts stays active. TCP listen socket stays listening.


> Multi-socket select() dispatch loops; managing listen queue depth, and dealing with non-blocking sockets (so that the select works) are also on the horizon for the modem implementers. And all of this to get a 1-hop discovery mechanism running - I wonder what the path length (libraries INCLUDED) for that is, just to avoid a single line of config (that being, the address and port of the other end of the connection)? 

I would not recommend to implement it this way.


> 
> The alternative I was shooting for, from the modem perspective, was along the lines of: 
> 1. Open a (blocking? non-blocking? don't care) UDP socket. 
> 2. Receive the discovery packet. Close that socket? Keep it open? Your call. 
> 3. Open a TCP socket. Issue connect(), from the info you got (and/or derived) from the discovery packet.
> 4. Flail away, sending and receiving actual DLEP protocol packets!

Looks complex to me.
Just use:
1. start simple DLEP discovery daemon, sending a UDP link-local multicast packet every now and then
2. start DLEP listener daemon
3. on TCP SYN, spawn a DLEP proces

For 3 there are other options. I’m pretty sure there will be a high performance open source DLEP implementations before RFC status. 


> 
> I'm beginning to think we need to pop the champagne corks and congratulate ourselves for (partially) designing something no one in their right mind would implement…

You are wrong. TCP servers on modems are implemented today. 


Teco


> 
> Stan
> 
> _______________________________________________
> manet-dlep-rg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/manet-dlep-rg