Re: [dccp] SOCK_DCCP

[Yoshifumi Nishida <nishida at csl.sony.co.jp>]

From: Ian McDonald <iam4 at cs.waikato.ac.nz>
Subject: Re: [dccp] SOCK_DCCP
Date: Mon, 08 Aug 2005 13:47:31 +1200
Message-ID: <42F6B9B3.9040800 at cs.waikato.ac.nz>

 > >>Hello,
 > >>I have one minor question.
 > >>Some or most? DCCP implementations use SOCK_DCCP for dccp sockets.
 > >>However, I found that some implementations already have another 
 > >>socket type which has similar concept: SOCK_CONN_DGRAM, although 
 > >>it has been used for ATM only and it is very rare to be seen.
 > >>Do we need two similar socket types?
 > > 
 > > 
 > > Further info about SOCK_CONN_DGRAM:
 > > 
 > > http://publib.boulder.ibm.com/infocenter/pseries/index.jsp?topic=/com.ibm.aix.doc/aixprggd/progcomc/skt_types.htm
 > > 
 > > SOCK_CONN_DGRAM:
 > > 
 > > Provides connection-oriented datagram service.  This type of socket
 > > supports the bidirectional flow of data, which is sequenced and
 > > unduplicated, but is not reliable. Because this is a connection-oriented
 > 
 > This is the reason we would need to be different I would suspect. DCCP
 > is sequenced, unduplicated and not reliable BUT it is two half
 > connections if data needs to flow both ways which is different to this.

Hmm, How the programming style would be different in this case?
Yes, DCCP has two logical half connections and if this means we have to do
listen() and accept() on both sides when data needs to flow both, the programming 
style will be very different. However, I don't think we need to do listen() on 
both sides. (I wonder why DCCP API doc doesn't exist...)

Thanks,
--
Yoshifumi Nishida
nishida at csl.sony.co.jp