[PATCH] Using TCP_NODELAY unconditionally

[Tobias Ringstrom]

On Mon, 21 Jan 2002, Tobias Ringstrom wrote:

> When running on the local interface on the Athlon, TCP_NODELAY gives much
> better performance than one overlapping request.  Increasing the number of
> overlapping requests to something large (10) achieves almost the same
> performance as with TCP_NODELAY.

I forgot to mention that this may very well be a problem with the Linux
Nagle implementation.  It would be very interesting to hear how other
OS:es behave.  Anyone?  Just apply the overlap patch from my last mail and
try different values of OVERLAPPING_REQUESTS (say 0-10).

/Tobias