On Sat, Oct 29, 2011 at 5:24 AM, David Miller [off-list ref] wrote:

From: Rick Jones <redacted>
Date: Fri, 28 Oct 2011 15:40:24 -0700

quoted

That would be loads faster, but won't that have issues with
granularity?

Frankly, I don't care.

For an obscure feature I don't even like to begin with, I refuse
to allow a multiply into a core code path.

I agree that there is no place for multiplication.

I think the best way to go right now is as follows:
* make TCP Delack params configurable per socket.
* count number of bytes instead of number of segments.

Although, I like more to have tcp_delack_thresh
measured in number of segments instead of number
of bytes I don't see yet a neat solution of how to avoid
the multiplication (shift count is not acceptable because it
limits the available number of segments).

Any comments?

thanks,
Daniel.