Re: [ns] NUMDUPACKS

[Lloyd Wood <l.wood@eim.surrey.ac.uk>]

On Wed, 15 Nov 2000, Arzad Alam Kherani wrote:

> 	The following line is from tcp.h:
> 
> #define NUMDUPACKS 1		/* normally 3, sometimes 1 */

recompiling ns every time you change this
should-have-been-const-or-better-bound
value is a real pain.

Ethan fixed that; see the thread starting:
http://www.isi.edu/nsnam/archive/ns-users/webarch/2000/msg03606.html

and I'd advise patching or upgrading your ns.

#define NUMDUPACKS 3            /* This is no longer used.  The variable */
                                /* numdupacks_ is used instead. */

> I wanted to know if there are real TCP implementations which have
> NUMDUPACKS set to values other than 3.

There's an internet-draft out (allman, floyd) talking about lowering
it to 2 in certain small window conditions.

http://www.normos.org/ietf/draft/draft-allman-tcp-lossrec-00.txt

No-one's looked at (or rather, published) the effects on large amounts
of traffic/flows of selectively raising the threshold to increase
tolerance for large amounts of misordering without loss when it's
experienced afaik; you could work on developing algorithms to do that.
It's much easier to have an algorithm change a bound variable than a
compiler definition, too.

(I spent some time fiddling with dupack thresholds and DV multipath
 routing to get misordering.)

L.

#define is _so_ 1978.

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>