On 2010-02-04, Sebastiano Pomata <[email protected]> wrote:
> Why on the FAQ (chapter 6) it says that tweaking
> net.inet.tcp.recvspace and
> net.inet.tcp.sendspace won't led to great improvements, while actually
> I got them?
$ cvs annotate faq6.html|grep very.few
Annotations for faq6.html
***************
1.234 (nick 01-May-06): Note that very few will see any benefit from
this.
This is less true now than when it was written, the average internet
connection has changed a bit since then - fast connections with a bit
of latency (especially if the server you're downloading from is at
some distance) are far more common than they used to be.
You could look at this as a value to estimate the tcp window size
(tcp.{send,recv}space) :-
1,048,576.00 byte/s minimum bandwidth on link (8Mbit/s)
* 0.05 s expected latency (50ms)
============
52,428.80 byte
Of course increasing this means that more kernel memory is used
for buffering.