Re: [shara] port randomization (draft-ymbk-aplusp-03)

[Rémi Després <[e-mail]>]

[e-mail]  -  le (m/j/a) 3/12/09 6:51 AM:
>   >-----Original Message-----
>   >From: [e-mail] [mailto:[e-mail]] On Behalf Of
>   >ext Dan Wing
>   
>   >Has consideration been given to having the PRR return only *one* port
>   >for each request, or to returning a list of port numbers which are
>   >not consecutive and are not a bit-pattern of ports?  These techniques
>   >would allow the PRR to distribute the requests randomly across the
>   >entire port range instead of within a block of ~100 (or whatever).
>
> This is exactly the intention of section 4 and 5 in http://www.ietf.org/internet-drafts/draft-bajko-pripaddrassign-01.txt
>
> What section 5 describes is a way to communicate a list of preallocated random ports to the client, in an indirect way.
>
>   
draft-bajko-pripaddrassign seems to me an excellent document to start with.

In www.nabble.com/FYI:-draft-despres-sam-02--enclosed-td22493319.html, I 
propose another method to avoid using consecutive port numbers on the 
global Internet.

If agreeable, I could propose a new section to deal with it in 
draft-bajko-pripaddrassign.

Parameters to be advertised in what could be sub-option 3, would be:
- the IPv4 address
- the dynamic port prefix
- the scrambling multiplier (e.g. 32 bits)
- the number of high order bits of the IPv4 address that must not be 
scrambled
 
This new sub-option is expected to be simpler than sub-options 1 and 2, 
and good enough.

In some respect it is also more powerful because it can  randomize not 
only port bits but also some lower bits of the IPv4 address: those that 
are after the IPv4 prefix of the multiplexing gateway (aka the PRR).

I look forward to discussing all this in SF.

Regards,

RD