Detour:  The Angry Triangle Years

Data Files

• sampleCountCDF_phase1.jpg
• sampleCountCDF_phase2.jpg

• comparisonOfStdDevToMeanLatencyForPathAB_logScale.jpg
• comparisonOfStdDevToMeanLatencyForPathAB_regularScale.jpg
• deviationsInMeanLatencyForDifferentRoutesForPathAB.jpg (directed paths; AB != BA)

• Why result loss rates are so high - the popularity contest. Reference counts on nodes appearing in desired alternate paths

• Comparison, Phase 1 vs Phase 2 Median Latency mag improve graphs, TT5MdLL
• Magnitude Improvement, Median Latency, PHASE ONE, best alt path wrt TT5MdLL

• Absolute Loss Improvement, best alt according to TT5MdLL
• Magnitude improvement, mean latency, best alt according to TT5MdLL
• Magnitude improvement, median latency, best alt according to TT5MdLL
• Magnitude Improvement, loss rate, best alt according to TT5MdLL
• Absolute improvement, median latency, best alt according to TT5MdLL

• Magnitude Improvment, Mean Loss, Instantaneous:Path vs. Best Alternate according to TT5MdLL
• Magnitude Improvment, Mean Latency, Instantaneous:Path vs. Best Alternate according to TT5MdLL
• Correlation : Latency Improvement Magnitude with Ngon Measurement Sample Count

VBNS paths removed, resampled to 50 min due to VBNS mixture paths. Now 602 total paths.
• Excel format of graphs below.
• Absolute Improvement, median latency, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Magnitude Improvement, median latency, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Absolute Improvement, Loss, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Magnitude Improvement, Loss, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Magnitude Improvement, Loss
• Magnitude Improvement, Median Latency
• Magnitude Improvement, Mean Latency
• Absolute Improvement, Loss
• Absolute Improvement, Median Latency
• Absolute Improvement, Mean Latency

Warts removed.
• Absolute Improvement, median latency, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Magnitude Improvement, median latency, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Absolute Improvement, Loss, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)
• Magnitude Improvement, Loss, best alt path wrt T5MdLL (Top 5 Median Latency, Lowest Loss)

• That wart removed, the graph all spiro-graph-y.
• An additional wart on the graph.

• MeanRTT / StdDev vs Loss, path AB
• StdDev RTT vs Loss, path AB
• Std. Dev / Mean RTT vs Loss, path AB

• Excel format of graphs below.
• Absolute Improvement for Loss, best alt. path wrt (mean r)root p
• Absolute Improvement for Loss, best alt. path wrt (median r)root p
• Absolute Improvement for mean latency, best alt. path wrt (mean r)root p
• Absolute Improvement for mean latency, best alt. path wrt (median r)root p
• Absolute Improvement for (mean r)root p
• Absolute Improvement for median latency, best alt path wrt (mean r)root p
• Absolute Improvement for median latency, best alt. path wrt (median r)root p
• Absolute Improvement for (median r)root p
• Magnitude Improvement for Loss, best alt. path wrt (mean r)root p
• Magnitude Improvement for Loss, best alt. path wrt (median r)root p
• Magnitude Improvement for mean latency, best alt. path wrt (mean r)root p
• Magnitude Improvement for mean latency, best alt. path wrt (median r)root p
• Magnitude Improvement for (mean r)root p
• Magnitude Improvement for median latency, best alt. path wrt (mean r)root p
• Magnitude Improvement for median latency, best alt. path wrt (median r)root p
• Magnitude Improvement for (median r)root p

• Excel format of 8 graphs below
• Absolute Improvement, Loss (best alt. for mean latency)
• Absolute Improvement, Loss (best alt for median Latency)
• Absolute Improvement, Mean Latency (best alt for Loss)
• Absolute Improvement, Median Latency, (best alt for Loss)
• Magnitude Improvement, Loss (best alt for Mean Latnecy)
• Magnitude Improvement, Loss (best alt for Median Latency)
• Magnitdue Improvement, Mean Latency (best alt for Loss)
• Magnitude Improvement, Median Latency (best alt for Loss)

• Excel format -- all graphs below. (more easily imported/played with in Powerpoint)
• Mean Latency, magnitude improvement of best alternate
• Median Latency, magnitude improvement of best alternate
• Mean Latency, absolute improvement of best alternate
• Median Latency, absolute improvement of best alternate
• Means: Path effect: magnitude improvement of latency vs. difference in hop count.
• Medians: Path effect: magnitude improvement of latency vs. difference in hop count.
• Packet loss effect graph -- hypothetical detour efficiency
• Relative Loss graph
• Absolute Loss graph

• Correlation between loss rate and number of loss samples.  Note that all of our high loss rates appear to be spurious.

Triangle Retreat Talk Slides

The following graphs have been generated at one point or another, archived here for the non-masses.

• 7/8/98 Priority List
• 7/7/98 The slime[-trail] graph for latencies. Series indices are GMT times, plus one hour (ie, Series 12 is GMT 11). Thus, 24 series, each one hour long.  PST is GMT - 8. The reason I'm not doing fraction of paths on the bottom is due to the sheer annoyance that I have with Excel right now; it'd take a while.  Top light-blue line, in both, is Series 9.  Bottom magenta line is Series 2. Paths graphed are {(abs lat)11-500, (mag lat)3-512}, of ~850 total (varied between ~825 and ~900, implying that some paths were not reachable at some times, or perhaps just low sampling.)
  absolute latencies, magnitude latencies
  
  
• 7/7/98 The original Xmas Trees; Hops vs Drops and Latency vs Drops.  This is simply for the path AB, not anything best, better, or best better.
  Hops vs Drops, Mean Latency vs Drops
  
  
• 7/7/98 Still using the June23 Dataset presented at the retreat; The following are the distributions of loss samples and latency samples used in the graphs generated.  There's definite xmas-treeing in the graphs, more so in the loss; this may be propogating over to the quantitative measurements (ie, the growth of one tree may be rooted in another).
  latency, drops
  
  
• 7/2/98 Correllation between router-sets; ie, given a path AB, how many other paths XY shared at least one interior node with it?
  Filters:
  1. If any one of {src=src,targ=targ,src=targ,targ=src} is true, then this match is uncounted.
  2. If the node is similar in the top 24 bits of the address to either the source or the target, this match is uncounted.
  1063 paths comprise the x axis.
  graph
  
  
• 6/25/98:Packet Loss as the triangle metric.  Here we see the packet loss probability for the path AB along-side the loss probability for the best better one-hop indirect path, where the probability for loss on the indirect link is calculated as p(ax) + (1-p(ax))*p(bx).
  graph
  
  
• The following are graphs of the magnitude of improvement to be obtained from traversing an alternate path.  Specifically, given start/end nodes AB, and an optional intermediary node X, they show the magnitude of improvement gained by traversing the path A->X->B instead of traversing A->B directly. 

  The cost metric is latency, and unless otherwise stated, the graphed values are the averages of the path round-trip latency over some time period (often all time), where the path AB is considered to be equivalent to the path BA, due to measuring the round-trip values.

  Small error:  The number 622 (of total paths measured) should be around 900 to 1100, in the graphs below.

  Further note:  Data files are in CSV format, suitable for most applications.

  1. • May 24-June 11, all data averaged, all times.
     • May 24-June 11, data > 1 in value, averaged, all times.
     • data set
  2. • May 24-June 11, 6am PST to 4pm PST vs 5pm PST to 2am PST.
     • Daytime data
     • Evening data
  3. • May 24-June 11, 6am PST to 4pm PST, vs. 9pm PST to 6am PST.
     • Daytime data
     • Nighttime Data

WRT the code, the initial parser examines the total data set via the Nexus implementation (slow), and spits out a file containing:

• The path under examination
• The best alternate path
• The average latency calculations, including standard deviation for all three.

• Initial Parser
• Secondary Parser

For the archivistically curious, the original measurement proposal is available from here.