Re: Question regarding the IEEE floating point format used in RFC 363 0

[Dave Katz <dkatz at JUNIPER.NET>]

On Jul 29, 2004, at 5:42 AM, Nelke, Jens wrote:

> Why weren't the values stored in unsigned long with the units
> Kilobits/s if
> I remember correct this kind of encoding is used in some of the MPLS TE
> MIBs.


The really short answer is that we did it the same way as IS-IS in
order to keep the drafts aligned.

The slightly longer answer is that 23 bits of mantissa gives roughly
seven significant figures of accuracy, which ought to be good enough
for most purposes.  It does not allow for taking 300 bps reservations
out of a 10 Gbps link, but that level of granularity was not
envisioned.  Also, IEEE floats are easy to work with in practice.

The problem with fixed point is that it's fixed point.  32 bits is
generally not enough to future-proof anything.  I suppose both OSPF and
ISIS could have been defined as 64 bit fixed-point integer bps, though
this wouldn't work for longwave submarine communications or something,
and it would have been viewed as overkill.

If it turns out that the world really needs more than seven orders of
magnitude of granularity in TE reservations, it would be simple enough
to create a new TLV that has the same semantics as the current scheme
but is formatted differently.