On Fri, 3 Jun 2005, Jonathan Stone wrote:
>>> How about wording like this:
>>>
>>>   The TCP implementation in
>>>   .Nx
>>>   includes NewReno (which subsumes the earlier TCP-Reno and
>>>   TCP-Tahoe versions), TCP Selective Acknowledgements (SACK,
>>>   RFC-2018, rfc-2883), and TCP-Westwood.
>>
>> OK
>
> Thanks.  We should probably say something about initial flight-size
> (rfc2414?) and flight-size (or cwnd size) after a congestion event (or
> if you prefer, the initial cwnd size, and cwnd size after congestion).
>
> One of the half-dozen things TCP-Vegas did that got better performance
> than Tahoe was an initial cwnd and post-drop cwnd of 2 segments versus
> 1 segment in TCP-Tahoe.  OTOH, I'd have to go UTSL to check exactly
> what we offer.
>
>>> followed by another sentence in the same paragraph, detailing which
>>> options are off by default and require sysctls to enable them?
>>
>> I prefer not to go near that. Our sysctl documentation is a serious mess,
>> and if someone wants to sort that out, that would be very great.
>> I.e. for a start, a description of all tcp related sysctls could be put
>> into tcp.4, and then tcp.4 be Xref'd from sysctl.[89], i.e. move
>> documentation of the sysctl knobs to the related topic.
>>
>> There seems to be quite a few of these knobs, of which I know 0. :(
>
>
> Fair enough. doing sysctl net.inet.tcp shows... well, to me it looks
> like it'd seem a tangled mess to your average sysadmin.

Yep. Would you be interested in updating tcp.4? :)


  - Hubert

-- 
Genervt von Viren und von Viren versandten Spam-Mails?  -> www.NetBSD.de
Fed up from viruses & virus-originated spam mail?       -> www.NetBSD.org