I wanted to plot a set of data by 3 dimensions.

I wanted to plot I/O read latency by MB/s throughput by number of concurrent readers. Seemed simple. Well it turns out there is no good way to do it in Excel. Sure Excel has 3d charts but, attention, the z axis is treated like rows and not values. For example

Note that the z axis, “users” had 3 values marked  on the axis. Those 3 values are 1,16 and 64. Notice that  16 is as far from 1 as 64 is from 16, ie the distance is not proportional to the value.

There is a free plug-in for Excel called Excel3Dscatterplot, but the data is hard to read, for example

Luckily R saves the day. With R there are a number of easy ways to graph 3d data.

Let’s take a data set that represents network I/O showing  the send size in KB (s_KB),  number of concurrent sender threads (thrds), the MB send throughput (s_MB/s) and then avg, min and max latency (mn_ms, avg_ms, max_ms).

to plot this in R, first start R up (see http://scs.math.yorku.ca/index.php/R:_Getting_started_with_R#Installing_R )

find out where R’s working directory is

   getwd()

go to a directory with my data and r files

list files

load and plot
   nio <- read.csv("perfibmd1_loopback_short.csv")

   nio
s_KB thrds mn_ms avg_ms max_ms  s_MB.s
1    NA     1  0.02   0.05   3.06   0.010
2    NA     8  0.04   0.12   5.27   0.024
3    NA    64  0.20   0.82  12.03   0.025
4     8     1  0.03   0.06   8.47  70.529
5     8     8  0.04   0.17  21.64 150.389
6     8    64  0.23   1.31  20.50  48.604
7    32     1  0.06   0.10   1.82 149.405
8    32     8  0.07   0.32  16.78 321.392
9    32    64  0.18   5.32 380.02  73.652
10  128     1  0.03   0.28   2.01 226.457
11  128     8  0.01   0.80  54.78 557.196
12  128    64  0.06  11.77  77.96 180.176
13 1024     1  0.01   1.49   5.76 335.587
14 1024     8  0.01   5.35 118.48 726.876
15 1024    64  0.02  40.50 221.59 714.162


library(scatterplot3d)
MBs=nio[,'s_MB.s']
threads=nio[,'thrds']
IOsize=nio[,'s_KB']
s3d <-scatterplot3d(IOsize,threads,MBs)


 

That’s a bit simple, but we can add some easy improvements. Here the dots are made solid, with color highlighting and a drop line to the bottom plane

x=nio[,'IOsize']
y=nio[,'threads']
z=nio[,'s_MB.s']
s <- scatterplot3d(x,y,z,xlab="IOsize", ylab="threads", zlab="MBs")
   for ( i in  1:5  )  {
     j=i*3
     p1 <- s$xyz.convert(x[j-2],y[j-2],z[j-2])
     p2 <- s$xyz.convert(x[j-1],y[j-1],z[j-1])
     p3 <- s$xyz.convert(x[j],y[j],z[j])
     segments(p1$x,p1$y,p2$x,p2$y,lwd=2,col=i)
     segments(p2$x,p2$y,p3$x,p3$y,lwd=2,col=i)
  }

Kyle Hailey
Uncategorized