[BXXPwg] Multiple sequence frames

Marshall Rose mrose+mtr.netnews@dbc.mtview.ca.us
Tue, 18 Jul 2000 01:17:32 -0700


we should probably add something to the tcpmapping document that explicitly
doesn't let a receiver reduce their window. my reading of the existing spec
says you can't do this because before you can send a SEQ you have to reserve
the buffer space, but maybe the phrase "accepts responsibility" means
different things to different people...

/mtr

----- Original Message -----
From: Sarveshwar Rao Duddu <duddu@vsnl.com>
To: Beep <bxxpwg@lists.invisible.net>
Cc: Marshall Rose <mrose+mtr.netnews@dbc.mtview.ca.us>
Sent: Monday, July 17, 2000 11:17
Subject: RE: [BXXPwg] Multiple sequence frames


>
> Oh :( I did not read the specs properly.
> But then there is another issue:
>
> C: SEQ 1 100 2048
> C: SEQ 1 100 1024
>
> That is, the client advertises a smaller window. When the first SEQ frame
is
> received, the server may find that the advertised window size is enough
for
> it to send a frame. Now the client has a *moral* responsibility to honor
> that frame. And it is not possible for it to do so because it does not
have
> enough memory.... The above sequence may happen if the client is having
only
> one buffer for both sending and receiving data, and just after it sent the
> first SEQ frame the application on the client side sent some REQ.
>
> If such a situation has to be avoided, there has to be two buffers for
each
> channel at both the peers. That pops up the resource utilization issue -
> what if the communication is such that requests are small and responses
are
> large, or vice - versa. Ok, this can be solved using buffers of unequal
> sizes, but then what happens if the peers reverse roles?
>
> Regards,
> Sarv
>
> is there any text in the tcpmapping document that makes you think
otherwise?
>
> /mtr
>
> ----- Original Message -----
> From: Sarveshwar Rao Duddu <duddu@vsnl.com>
> To: Beep <bxxpwg@lists.invisible.net>
> Sent: Monday, July 17, 2000 10:22
> Subject: [BXXPwg] Multiple sequence frames
>
>
> > Hi,
> >
> > I hope there is no requirement that a seq frame be sent only when a
> REQ/RSP
> > is received. That is, the receiver may send multiple SEQ frames even
when
> > the sender is idle. This would make the receiver possible to know the
> > current status of the buffers at the receiver side. I think this is
> > possible, and if yes, should be made part of the specs clearly. Am I
> right?
> >
> > Regards,
> > Sarv
> >
> >
> > _______________________________________________
> > BXXPwg mailing list
> > BXXPwg@lists.invisible.net
> > http://lists.invisible.net/mailman/listinfo/bxxpwg
> >
>
>