Diagnostic Tools

Thu Jun 6 17:31:17 UTC 2002

      Jane Pawlukiewicz wrote:
    > Ping and traceroute give me a ton of data. I was thinking of something
    > that takes that data and turns it into the bottom line. Where is the
    > problem, when did it start, all the good stuff.

I think that's called Sean Donelan.  :-)

To give you a serious answer, though, there are a few reasons why this is
a problem that smart software developers are leery of tackling.  Two big
ones are:

- What to measure?  Loss, latency, jitter and path length and changes are
  obvious metrics, but where do you measure to and from?  Do you measure
  from the desktop machine of whoever buys your software, or do you
  measure from somewhere or some large set of somewheres which might be
  more representative of the Internet overall, at the risk of being less
  representative of the customer themselves?  Do you measure to some set
  of generic frequently-viewed web sites, although this is likely to
  annoy the proprietors of those sites, if the tool becomes popular?  Or
  to some set of routers within the backbone infrastructure, although
  someone may get wise and put them on private addresses or cause them to
  stop wasting cycles responding to your tool?  Is there even a right
  answer to this?  It may be that one size doesn't fit all.

- If you know what you want to measure to and from, can you observe the
  path in both directions?  In order to do either active or passive
  measurement of a path, you have to have devices in that path, and a path
  is generally uni-directional for at least a portion of its length.  That
  is, the forward and reverse directions pass through different equipment
  across different links, utilize capacity differentially in each
  direction, and share available capacity with other flows which are
  utilizing it differentially as well.  If you think about what this
  means, the unfortunate conclusion that most people reach is that even if
  one were able to distribute thousands of probes throughout the Internet,
  one would still only be able to measure a _tiny_ portion of the paths,
  and the portion is tiny enough that it may not be sufficient to
  extrapolate any useful statistics from.

                           -Bill