Skip to Content.
Sympa Menu

transport - Re: [transport] Minutes for 2005-03-04 transport teleconference

Subject: Transport protocols and bulk file transfer

List archive

Re: [transport] Minutes for 2005-03-04 transport teleconference


Chronological Thread 
  • From: Yunhong Gu <>
  • To:
  • Subject: Re: [transport] Minutes for 2005-03-04 transport teleconference
  • Date: Fri, 4 Mar 2005 15:22:07 -0600 (CST)


Hi, all,
I have several questions about this max datagram size.

>
> - Max library datagram size can be large; e.g., an unsigned 32-bit
> quantity.
>

I doubt if it makes sense to allow unlimited (datagram) message size.
People probably wouldn't call a 100MB buffer a "message". In addition, if
a "message" contains 1000 packets/segments, the probability of losing
one of the packets (including its retransmissions) is very high, thus few
of such large messages will be delivered successfully (in a timely
manner).

second question, will the messages be delivered in order?

third question, if part of a message is lost, will it be discarded or
will certain application may still be interested in the damage message?
i.e., what is the semantics of recv() call on SOCK_DGRAM? Is it (x_recv)
transparent to damage message?

For example, if in the receiver buffer, there are the following messages
(where a 'd' means the message is damaged):

1 2(d) 3 5 4

which one of the following sequences should x_recv() return?

1 3 5 4
1 3 4 5
1 2 3 5 4
1 2 3 4 5

Thanks,
Gu



Archive powered by MHonArc 2.6.16.

Top of Page