On Sun, May 06, 2001 at 04:09:56PM +0100, Adam Langley wrote:
> On Sun, May 06, 2001 at 04:36:47PM +0200, Oskar Sandberg wrote:
> > No, we can't have this. Such a system means that a whole part has to
> > received before it can be sent out. That makes sense of FCP, but in FNP we
> > need to start returning the data as soon as possible (or we end up in
> > another situation where we have to wait for an unspecified amount of
> > time - if you have 15 hops and a part takes 5 seconds, then that will be
> > an extra 75 seconds at the client before the data comes).
>
> Well, Tavin can't be bothered to code it - so the discussion is
> academic. But you don't have to wait for the whole part - send chunks
> as you get them. If the block fails to verify - send 0x0000 and
> restart. The recving node just checks the part once it has it - if the
> upstream node said everthing was good and it isn't - write 0x0000 to
> the downstream and restart.
How is this any different from what we have? If the upward stram fails in
the middle of one of these chunks you will still have to nullpad it...
> > Padding to the end of the block is not hard. Stop complaining.
>
> No, it's not hard. It's just a pain.
The only way to not have to pad is if you KNOW you will be able to send
the whole block. The only way to KNOW that is if you already have the
whole block.
--
'DeCSS would be fine. Where is it?'
'Here,' Montag touched his head.
'Ah,' Granger smiled and nodded.
Oskar Sandberg
[EMAIL PROTECTED]
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
