[e2e] bytes vs packets

Fernando Gont fernando at gont.com.ar
Tue Jul 29 12:14:58 PDT 2003

At 21:54 28/07/2003 -0400, Craig Partridge wrote:

> >While discussing the third reason (3), he states "It was not clear if this
>advantage would be important; it turned out to be critical."
> >Wasn't any sending-side solution (such as Nagle's algorithm) considered at
> >that time?
>Nagle's is a rule for combining data before creating a TCP segment.

Yes, I do know.

>The issue that Dave refers to is situations such as when you've sent
>two segments and on retransmission send one segment containing the
>data of both.  (More broadly, the text you quote seems to reflect TCP pre
>the invention of IP).

Yes, but my question is why was repacketization considered so *critical*.

As I mentioned above, I wonder if some sending-side solution was 
considered, instead of considering the problem Clark describes (see bellow) 
as one of the driving forces for choosing to acknowledge bytes instead of 

Clark describes the problem as:

"Systems such as UNIX which have a internal communication model based on 
single character interactions often send many packets with one byte of data 
in them. (One might argue from a network perspective that this behavior is 
silly, but it was a reality, and a necessity for interactive remote login.) 
It was often observed that such a host could produce a flood of packets 
with one byte of data, which would arrive much faster than a slow host 
could process them. The result is lost packets and retransmission.
If the retransmission was of the original packets, the same problem would 
repeat on every retransmission, with a performance impact so intolerable as 
to prevent operation. But since the bytes were gathered into one packet for 
retransmission, the retransmission ocurred in a much more effective way 
which permited practical operation."

My point is that with a sending-side solution such as Nagle's algorithm, 
the "flood of packets with one byte of data" would never happen. Actually, 
those packets probably would never need to be retransmitted, as they would 
be sent efficiently from the very first time.

Fernando Gont
e-mail: fernando at gont.com.ar || fgont at acm.org

More information about the end2end-interest mailing list