'tis strange, sir, tis very strange;
That is the brief and the tedious of it
-- William Shakepeare, All's Well that Ends Well
we are experimenting with methodologies for assessing and
visualizing delay and congestion indicators along paths to a
specified set of destinations. this graph below is the
result of a half an hour sample of data collection
taken at 1600EST on Tuesday June 18, 1996,
from the NY node of the vbns, to a set of 24 destination hosts.
The data collection progresses in two phases:
we first gather topology data incrementally, fanning out from
our source host toward each of the destinations in our set.
As we build the toplogy map, we keep one echo request in flight
to a randomly chosen hop in our current link set.
The shape of each node reflects its role in the infrastructure:
hexagons correspond to routers located at NAPs or other
interconnection points; ovals correspond to routers belonging
to transit backbones; and squares to routers of local or
regional providers. The color of each node represents the
variance in round trip time of these echo requests, where
red is high and blue is low.
After running the data collection tool for several days, we were able to build an animation using a separate frame for each half hour worth of samples. Unsurprisingly, the variance is higher during this snapshot then it is during evening or weekend collection intervals, particularly noticeable at the NAPs, and also at the periphery of some of the regional providers.
Even more interesting in the graph is the meshes of paths to given destinations, which reflect route flapping and occurred often during the duration of our data collection. The upper right quadrant of the graph, which we zoom in on below, indicates a remarkable degree of routing instability, in this case localized in the northeastern US.
Delays sampled across 30 minutes at 1600EST
on 18 June 1996 from it.cache.nlanr.net
(Ithaca node of the vBNS) to
several destination hosts
We use this small sample of data to explore
techniques assessing Internet connectivity from
a macroscopic perspective in near real-time.
Such tools could be attractive those trying to
Such tools could be attractive those trying to
operate, engineer, and characterize performance
as the Internet scales up in complexity.