Below.
On 21/01/2013 10:55 AM, Ole Troan wrote:
Tom,
Section 5.1.3 calls for the (optional) provisioning of a range of
excluded ports as part of a mapping rule. As my notes in Decemeber
hopefully made clear, the key issue is really whether system ports
(0-1023) are to be excluded or not. How many actually get excluded
is a function of block size, or of the sharing ratio if a = 0. The
default value of 4096 currently given in Section 5.1.3 is the block
size implied by an offset a = 4.
I would think an operator would not want to exclude any more ports
than necessary. Hence what needs to be provisioned is simply a
Boolean indication that system ports are to be excluded or not. If
they are to be excluded, then:
- for a = 0, as many PSIDs as necessary beginning with PSID = 0
MUST NOT be used because the corresponding port ranges lie within
the range 0-1023.
- for 1 <= a <= 6, the first block MUST NOT be used, because it
contains the system port range. The corresponding number of
excluded ports (including the system ports) ranges from 32768 for a
= 1, down to exactly 1024 for a = 6.
- a > 6, more than one block has to be excluded to make up the
complete range of 1024 ports.
The above bullets could be part of the explanation of the algorithm
in Section 5.1. Section 5.1.3 should change only by replacing
"excluded ports" by "whether system ports are excluded" and the
default value of that to "yes".
wouldn't it then be simpler to require that a MUST be either 0 or 6?
(or only 6). the discussion we've had earlier was that the benefit of
being able to easily recognise the PSID in a port, made up for the
lost 3K ports. which is what lead to an a=4.
cheers, Ole
The trouble is, the value of a is related to the number of blocks by the
formula
Number of blocks = 2^a
The number of blocks determines how many port ranges are allocated to
each MAP CE (i.e. with system ports and hence the first block excluded,
a = 4 implies that each sharing MAP CE is allocated 15 separate ranges
of ports). I think you want to retain the operator's flexibility to
allocate fewer blocks or, in the interest of reducing the chance that an
attacker can guess the next port allocated, more blocks. Of course, the
operators in the WG may disagree.
Tom
_______________________________________________
Softwires mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/softwires
