[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ippm] Review of: draft-ietf-ippm-reordering-12.txt

Hi Bert,

Thanks for your review and questions.
Please see my take on the answers, below.
hope this helps,
Al

At 05:55 AM 4/21/2006, Wijnen, Bert (Bert) wrote:

...
Looks pretty good.

I have one question that maybe the authors or other WG members can answer
for me and that is:

  In section 4.5, it seems to allow for using msg sequence numbers OR
  units of time (without even having defined what the unit is).

So I wonder how this definitions specifies an exact metric. The metric would
not be comparable from one to the other measurement if one of them uses
msg sequence numbers, while the other uses "units of time". Even if two of
them use "units of time" but different units (e.g. micro seconds vs milliseconds) 
even then they would not be comparable.


I'll tackle the issue of different time resolutions below.

In the case of the Gap metrics of Section 4.5, all the parameters
necessary to compute the Gap (in units of msg numbers) and
GapTime (in units of time) are mandatory, inherited from earlier
metrics (as 4.5.2 states).

If one reports a metric from this section, then the Gap metric is
mandatory, while the GapTime is optional:

      "Gaps may also be expressed in time:" (equation follows)

So, I don't think we should run into a problem when two different
testers report metrics that are compliant with section 4.5.  They
should be able to compare the Gap (msg number-based) metric, at least.



Was it not the goal of IPPM to define EXACT metrics, so that results of
two different tests/measurments could be compared?


None of the current IPPM metric RFCs mandate a resolution for time stamps.
The IPPM Framework RFC 2330 treats Clock Issues in Section 10,
and simply defines the term "resolution" as
"the smallest unit by which the clock's time is updated."

The One-way Delay RFC 2679 even allows for different resolutions
to be used on the Source and Destination clocks (from 3.7.1):

   +  The resolution of a clock adds to uncertainty about any time
      measured with it.  Thus, if the source clock has a resolution of
      10 msec, then this adds 10 msec of uncertainty to any time value
      measured with it.  We will denote the resolution of the source
      clock and the destination clock as Rsource and Rdest,
      respectively.


So, I believe we have specified the metric definitions "exactly",
or without ambiguity.  But the degree to which measurements
from different implementations will be comparable depends on
the details of each implementation, including the time stamp resolution
and other sources of error or uncertainty (such as time accuracy
and skew).