[sctp-dev] Using SctpMultiChannel

[Johan Lübcke]

A "connect" would be preferable. Another way be to get similar results would be if it was possible to do a send with a zero-sized buffer just to get the connection.

In my case we ended up with the multichannel due to our need use the same port and address for both incoming and outgoing connections.

  / Johan



11 aug 2011 kl. 15.30 skrev Chris:

> I see now that my reading comprehension isn't so good. So the multi-channel does create new associations also for incoming connections :)
> 
> But what I think I would really be after is a connect() method on SctpServerChannel that spawned a new SctpChannel. If so I wouldn't have to go to the multichannel and its somewhat different way of working.
> 
> BR,
>  Chris
> 
> --- On Thu, 8/11/11, Chris <devnullaccount at yahoo.se> wrote:
> 
>> From: Chris <devnullaccount at yahoo.se>
>> Subject: Re: [sctp-dev] Using SctpMultiChannel
>> To: "Johan Lübcke" <johan.lubcke at symsoft.com>, "Chris Hegarty" <chris.hegarty at oracle.com>
>> Cc: "sctp-dev at openjdk.java.net" <sctp-dev at openjdk.java.net>
>> Date: Thursday, August 11, 2011, 3:13 PM
>> Hi Chris and Johan
>> 
>> big thanks for the replies. 
>> 
>> I guess I could try send(), but I won't have any data to
>> send initially but would like to have the conneciton up so
>> it sounds a bit messy compared to my current code. But with
>> lack of other options, I'll try it.
>> Would the association change be a COMM_UP event with a
>> newly-created Association? And I guess that means the
>> "attachment" will be null when reaching the handler... 
>> 
>> Chris Hegarty said:
>>> It would appear that you want SctpChannel semantics
>> but not
>>> the overhead 
>>> of creating a new socket for each association. But
>> once
>>> branched the 
>>> returned SctpChannel will be assigned its own socket.
>>> 
>> 
>> Does it? As far as I understood RFC4960, you use the same
>> *socket* (since I'm using the same local IP/port) that
>> handles multiple different associations. So I'd like a new
>> association, which I guess is an SctpChannel but still bound
>> to the same socket. But maybe that's what you mean and I'm
>> just twisting words :)
>> 
>> I'm very surprised by the difference between the
>> SctpServerChannel and SctpMultiChannel since they are very
>> similar in what they do. Could I pervert the
>> SctpMultiChannel to do channel.register(OP_ACCEPT) and
>> channel.accept() in a similar way as to using "send()" to
>> create a connect()? 
>> I will need to run a single socket with that can do both
>> accept() and connect() on the same port, and currently I'm
>> not sure of which of these classes supports that.
>> 
>> BR,
>>   Chris
>> 
>> 
>> 
>> --- On Thu, 8/11/11, Chris Hegarty <chris.hegarty at oracle.com>
>> wrote:
>> 
>>> From: Chris Hegarty <chris.hegarty at oracle.com>
>>> Subject: Re: [sctp-dev] Using SctpMultiChannel
>>> To: "Johan Lübcke" <johan.lubcke at symsoft.com>
>>> Cc: "Chris" <devnullaccount at yahoo.se>,
>> "sctp-dev at openjdk.java.net"
>> <sctp-dev at openjdk.java.net>
>>> Date: Thursday, August 11, 2011, 2:59 PM
>>> 
>>> 
>>> On 08/11/11 01:25 PM, Johan Lübcke wrote:
>>>> I had the same problem and I ended up doing a
>> "send"
>>> call to the remote address. That way I get a
>>> AssociationChangeNotification with an association id
>> that
>>> can be passed to "branch".
>>> 
>>> Yes, this would be the best way to achieve what you
>> are
>>> looking for.
>>> 
>>> SctpMultiChannel is a little like a datagram socket,
>> in
>>> that it can send 
>>> and receive messages to/from remote peers without
>>> explicitly setting up 
>>> an association, i.e. no connect is required. The
>> class
>>> description and 
>>> send method describe this:
>>> 
>>>    "The channel will implicitly setup a new
>>> association whenever it
>>>     sends or receives a message from a remote
>>> peer if there is not
>>>     already an association with that peer."
>>> 
>>> It would appear that you want SctpChannel semantics
>> but not
>>> the overhead 
>>> of creating a new socket for each association. But
>> once
>>> branched the 
>>> returned SctpChannel will be assigned its own socket.
>>> 
>>> -Chris.
>>> 
>>> 
>>>> This seems a bit awkward but it works.
>> (Especially
>>> since I need to have some actual data to send already
>> when I
>>> want to connect to the remote side). Doing a "connect"
>> would
>>> have been more natural.
>>> 
>>> 
>>> 
>>>> 
>>>> If there is a better way I am equally
>> interested!
>>>> 
>>>>     / Johan
>>>> 
>>>> 
>>>> 11 aug 2011 kl. 14.03 skrev Chris:
>>>> 
>>>>> Hi,
>>>>> 
>>>>> I'm trying to use SctpMultiChannel to get
>> multiple
>>> associations on the same socket (same local port/IP
>> but
>>> different remote endpoints), but I can't for the life
>> of me
>>> figure out the API (http://openjdk.java.net/projects/sctp/javadoc/com/sun/nio/sctp/SctpMultiChannel.html)
>>>>> 
>>>>> I get that I should do
>> SctpMultiChannel.open() to
>>> create an instance and that I should bind it to my
>> local
>>> address(es).
>>>>> But then I want to use something like
>> connect()
>>> (like on a regular SctpChannel), but there's no such
>> method,
>>> or anything similar as far as I can see. And the only
>> way to
>>> get a regular SctpChannel out of this class seems to
>> be
>>> branch(Accociation), but that's not either an option
>> since
>>> I  need to have an association to call that.
>>>>> 
>>>>> What am I missing?
>>>>> 
>>>>> TIA,
>>>>>    Chris
>>>> 
>>> 
>> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 881 bytes
Desc: =?iso-8859-1?Q?Detta_=E4r_en_digitalt_signerad_del_av_brevet?=
Url : http://mail.openjdk.java.net/pipermail/sctp-dev/attachments/20110811/243aa851/attachment-0001.bin