[Roll] [roll] #18: Numeric Ranges in Routing Metric

["roll issue tracker" <trac at tools.ietf.org>]

#18: Numeric Ranges in Routing Metric
--------------------------------+-------------------------------------------
 Reporter:  jpv at …               |       Owner:  jpv at …        
     Type:  defect              |      Status:  new          
 Priority:  minor               |   Milestone:               
Component:  routing-metrics     |     Version:               
 Severity:  Active WG Document  |    Keywords:               
--------------------------------+-------------------------------------------
 This ticket is related to the Numeric Ranges in Routing Metric.

 Here is the original thread:

 I mentioned it during the roll meeting Tuesday, but I think it's probably
 worth documenting the thought in a little more technical detail.  The
 routing metrics right now include 32-bit floating point numbers for
 latency and throughput.  Since these numbers are much more likely to be
 added or compared rather than multiplied, fixed point rather than floating
 point is likely to be a better choice of representation.  For example, for
 normalized positive 32-bit IEEE 754 numbers, the maximum number is about
 3.4e38 and the minimum is about 1.2e-38.  For reference, the age of
 universe in milliseconds (taking the high end of NASA's most recent
 estimate) is only 4.4e20 and the time it takes to travel the distance
 across an electron (Lorentz diameter for you physics heads) is about
 1.9e-20 ms.  A negative latency would seem to imply that packets appear
 before they're sent which is probably not practical.  If instead, latency
 were a 32-bit unsigned integer in units of milliseconds, we'd still be
 able to go from 0 or 1 milliseconds up to almost 1200 hours if my math is
 right.  That should be plenty for any practical network.  I haven't done
 so but I suspect that we could do the same thing with throughput.

-- 
Ticket URL: <http://trac.tools.ietf.org/wg/roll/trac/ticket/18>
roll <http://tools.ietf.org/wg/roll/>