Importance of visualizing outliers

This rather busy figure shows the cache miss rate for chains of different numbers of network functions (NF) when sending 3 different sizes of packets (64, 500, and 1500 bytes) - using two different data plane implementations: Open vSwitch (OVS) and netmap/VALE into a chain of NFs as the length of the chain increases. Curves were fit to the data and the equations for these curves are shown below the figure. The equations are also shown in an HTML table below the figure. Note that the details of the figures and equations do not matter in this presentation, the focus is on how the data is plotted and the presence of the outlines.

Cache miss rates as a function of the length of a chain of network functions
Cache miss rates as a function of the length of the network function chain. Note that the vertical axis is logarithmic in this and the following plots.
Table showing the curves fit to the experimental data
method Size in bytes Equations R2 value
OVS 64 bytes y = 0.078 *L1_Miss_Rate + 0.0023 R2 = 0.99
OVS 64 bytes y = 0.00020 *L3_Miss_Rate + 0.0000138 R2 = 0.99
OVS 500 bytes y = 0.078 *L1_Miss_Rate + 0.0003 R2 = 0.99
OVS 500 bytes y = 0.00026 *L3_Miss_Rate + 0.0000021 R2 = 0.99
OVS 1500 bytes y = 0.079 *L1_Miss_Rate + 0.0023 R2 = 0.99
OVS 1500 bytes y = 0.00028 *L3_Miss_Rate + 0.0000006 R2 = 0.99
VALE 64 bytes y = 0.083 *L1_Miss_Rate + 0.0013 R2 = 1.00
VALE 64 bytes y = 0.00248 *L3_Miss_Rate + 0.0000373 R2 = 0.62
VALE 500 bytes y = 0.086 *L1_Miss_Rate + 0.0002 R2 = 1.0
VALE 500 bytes y = 0.00050 *L3_Miss_Rate + 0.0000063 R2 = 0.60
VALE 1500 bytes y = 0.088 *L1_Miss_Rate + 0.0001 R2 = 1.0
VALE 1500 bytes y = 0.00048 *L3_Miss_Rate + 0.0000019 R2 = 0.53

 

Data provided by Georgios Katsikas who was a doctoral student at Network Systems Lab (NSL), CoS at the time when this work was done - these data are shown on this and the following pages appear with his permission.

Transcript

This data was provided by Georgios Katsikas, a doctoral student at the network systems lab here at KTH [at the time when this work was done]. And he plots it and then tries to fit some curves to it.  And here we can see the data showing which colors correspond to which algorithms and which cache (the layer one cache and the layer three cache) but if we zoom in a little bit