--nextPart4299324.dJvKr78ukf
Content-Type: text/plain;
  charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline

On Saturday 23 October 2004 19:16, Jeffrey Hsu wrote:
> A philosophical point that loopback is supposed to faithfully mirror
> the same processing done for physical interfaces, else there's a whole
> lot more you can cut out, like TCP retransmissions and TCP out-of-order
> reassembly.

The "header prediction" path should handle a lot of this.  A couple of othe=
r=20
things that might make sense:

1) Disable time stamps.  We're using the same reference clock and the same=
=20
scheduler for retransmissions, so is there any point in bothering?

2) Disable slow-start.  This would improve performance in certain cases, bu=
t=20
it might risk exacerbating problems in low-memory situations.  It would nee=
d=20
some testing.

--nextPart4299324.dJvKr78ukf
Content-Type: application/pgp-signature

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (NetBSD)

iD8DBQBBesOq5ECq/rI19aIRAkAdAKCvUiKcx3+rjIiOTS+te5UbIVrOjgCgq7qW
hgA8l+5mMuWUNNLIiVwe8Ik=
=m1VO
-----END PGP SIGNATURE-----

--nextPart4299324.dJvKr78ukf--