On 02/ 6/10 12:22 AM, Darren J Moffat wrote:
> On 05/02/2010 15:56, Sebastien Roy wrote:
>> I'm submitting this self-reviewed case on behalf of Kacheong Poon as
>> closed-approved automatic. The release binding is Patch, and the
>> stability level TCP_INIT_CWND is Committed.
>>
>> In 2004/838, the TCP socket option, TCP_INIT_CWND, was introduced.
>> This option controls the initial congestion window (cwnd) used in
>> a TCP connection. The maximum value allowed is 8. This case is
>> to increase this maximum to 16. With today's network technologies,
>> this is a safe value to use in many environments. There is also
>> an on-going effort in IETF to increase the default initial cwnd.
>> This case does not change the default used in Solaris. An app
>> must still use the socket option to change the initial cwnd.
>
> Where does 16 come from ? Why not 32 or 64 ?
It is 2^4 so it must be a good number ;-) A burst of 32 TCP full
MSS size segments is probably not OK in most environments today.
Is 16 OK? Probably and there are already network equipments
using that value as the initial cwnd. And some customers want that.
The on-going IETF effort is to increase the default to 10. With a
little leeway, 16 seems to be good enough for now. If the need to
have a higher value comes, we can increase that.
--
K. Poon.
kacheong.poon at sun.com
