The RFC1323 just explain a mecanism to increase the maximum value of the
window.
I think you are right, the scale factor will be 9x in your case.
In the real world (real implementation :-), there is a strong impact of the
TCP configuration you are using on the choice of the scale factor,
especially from the socket size. I will explain what happend with a Linux
2.3 TCP. There some /proc files you can use to setup the default/maximum
value of the send/recv socket buffer size. During the setup of the
connection the sender will look at his socket buffer size and then compute
the biggest scale factor possible (the TCP linux implementation roughly
assume that half of the buffer will be use by TCP data, it may sound strange
but it is due to the socket buffer structure) and send this factor in the
first segment. The receiver does exactly the same with his own buffer and
send his value to the sender. the smallest value between the sender and the
receiver value will be used during the connection.
Of course, if you don't specify a special value for the size of the socket
buffer size, the default value will be used.
Hope it may help
Cheers
julien
> --Original Message-----
> From: Greg Otto [SMTP:[email protected]]
> Sent: Tuesday, November 30, 1999 8:43 PM
> To: tcpsat
> Subject: Selection of Scaling Factor
>
> When using RFC1323 with Window scaling, how is the scale value selected?
> My
> guess is it is based on the smallest scale (left shifts) required for a
> 64K
> to be greater than the configured maximum.
>
> For example, if a WINDOW/HIWAT of 18,000,000 bytes is selected, then the
> scale would be 512 (275 rounded up to the nearest multiple of 12) or a
> left
> shift of 9x.
>
> Is this correct or what is a resource which provides a better explanation?
>
>
> Thanks,
> GReg
>
>
> Greg Otto voice: 713-995-4778 x102
> New Frontier Consulting, Inc. fax: 281-754-9170
> Houston, Texas
> http://www.newf.com email: [email protected]
>
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:58 EST