Re: [patch] bug fix & TCP networking performance improvements

[Greg Troxel <gdt%ir.bbn.com@localhost>]

David Young <dyoung%pobox.com@localhost> writes:

> On Wed, Apr 20, 2011 at 01:57:09PM +0200, Matthias Drochner wrote:
>> 
>> tls%panix.com@localhost said:
>> > I don't see any such change in our patch.  Could you point it out?
>> 
>> Here (in your devel-vtw-diff1 file):
>> 
>> @@ -3231,9 +3492,9 @@ tcp_xmit_timer(struct tcpcb *tp, uint32_
>>                  * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
>>                  * point).  Adjust rtt to origin 0.
>>                  */
>> -               delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT);
>> +               delta = (rtt << 0) - (tp->t_srtt >> TCP_RTT_SHIFT);
>>                 if ((tp->t_srtt += delta) <= 0)
>> -                       tp->t_srtt = 1 << 2;
>> +                       tp->t_srtt = 1; // 1 << 2;
>>                 /*
>>                  * We accumulate a smoothed rtt variance (actually, a
>>                  * smoothed mean difference), then set the retransmit
>
> I have committed vestigial time-wait and maximum segment lifetime
> truncation.  I left the RTT changes out.  Do people agree that the
> changes are correct?

They are not obviously correct, and I think it's more likely than not
that they aren't correct.  Have you looked at them relative to the
comment enhancements I recently committed?