[MINA] Unbound streams

[Jose Alberto Fernandez <ja...@cellectivity.com>]

Hi I have been thinking about the different solutions we have for
Streams and really none of them really convince me.

They all require that one pre-computes the size of the stream before you
send anything, which means in most cases in memory buffering. This means
that for example, to send a Serialized object, one need to serialize it
into a buffer and then send it prefixed by the size.

 

One alternative to this would be to define a small protocol that can be
used to encapsulate an unbound stream in the middle of a transmission.
Something like:

 

1)       Sender transmits as part of its protocol a
BEGIN_STREAM(<end-mark>) header. <end-mark> is a sequence guaranteed not
to be part of the following stream content.

2)       Receiver has installed a filter that detects the BEGIN_STREAM
header and provides the receiver with an InputStream instance to be
consumed (by a separate thread, perhaps).

3)       Sender commences transmition of the stream content and ends
with <end-mark>.

4)       Receiver's filter sends all input via the InputStream instance
until it recognizes <end-mark> and closes the stream at that point.

5)       Any further inputs sent/received will be processed in the
regular manner.

 

You can see this is inspired on UNIX's "<< EOF" shell syntax. This will
allow us to use, for example, regular object serialization without the
need for buffering. I do not think it should be too difficult to come up
with some natural <end-mark> given the particular nature of the stream. 

 

What do you all think? How feasible is to do this in MINA?

 

Jose Alberto

 

Re: [MINA] Unbound streams

[Trustin Lee <tr...@gmail.com>]

Hi Jose,

2006/1/18, Jose Alberto Fernandez <ja...@cellectivity.com>:
>
>  Hi I have been thinking about the different solutions we have for Streams
> and really none of them really convince me.
>
> They all require that one pre-computes the size of the stream before you
> send anything, which means in most cases in memory buffering. This means
> that for example, to send a Serialized object, one need to serialize it into
> a buffer and then send it prefixed by the size.
>
>
>
> One alternative to this would be to define a small protocol that can be
> used to encapsulate an unbound stream in the middle of a transmission.
> Something like:
>
>
>
> 1)       Sender transmits as part of its protocol a
> BEGIN_STREAM(<end-mark>) header. <end-mark> is a sequence guaranteed not to
> be part of the following stream content.
>
> 2)       Receiver has installed a filter that detects the BEGIN_STREAM
> header and provides the receiver with an InputStream instance to be consumed
> (by a separate thread, perhaps).
>
> 3)       Sender commences transmition of the stream content and ends with
> <end-mark>.
>
> 4)       Receiver's filter sends all input via the InputStream instance
> until it recognizes <end-mark> and closes the stream at that point.
>
> 5)       Any further inputs sent/received will be processed in the regular
> manner.
>
>
>
> You can see this is inspired on UNIX's "<< EOF" shell syntax. This will
> allow us to use, for example, regular object serialization without the need
> for buffering. I do not think it should be too difficult to come up with
> some natural <end-mark> given the particular nature of the stream.
>
>
>
> What do you all think? How feasible is to do this in MINA?
>

Sure.  You can do that by implementing a state machine in your decoder.
You'll have to remember which field in a message you're reading now, and
reset the status when you reach the EOM (End of Message).  The reason people
uses prefixed data structure is because it is very easy to implement. :)

HTH,
Trustin
--
what we call human nature is actually human habit
--
http://gleamynode.net/
PGP Key ID: 0x854B996C