Originally posted by Vadim Tkachenko

In my previous post, I finished with a graph with unstable results. Now let's explore some different ways to present those results. I enjoy working with R, and though I'm not even close to being proficient in it, I want to share some graphs you can build with R + ggplot2.

The conditions of the benchmark are the same as in the previous post, with the difference being that there are results for 4 and 16 tables cases running MySQL 5.5.20.

Let me remind you how I do measurements. I run benchmarks for 1 hours, with measurements every 10 seconds.
So we have 360 points – metrics.

If we draw them all, it will look like:

I will also show my R code:

m <- ggplot(dv.ver,
            aes(x = sec, Throughput, color=factor(Tables)))
m + geom_point()

The previous graph is not very representative, so we may add some lines to see a trend.

m + geom_point() + geom_line()

This looks better, but still you may have hard time answering: which case shows the better throughput? what number we should take as the final result?

Jitter graph may help:

m <- ggplot(dv.ver,
            aes(x = factor(Tables), Throughput, color=factor(Tables)))
m + geom_jitter(alpha=0.75)

With jitter we see some dense areas, which shows "most likely" throughput.

So let's build density graphs:

m <- ggplot(dd,
            aes(x = Throughput,fill=factor(Tables)))
m+geom_density(alpha = 0.7)

or

m+geom_density(alpha = 0.7)+facet_wrap(~Tables,ncol=1)

In these graphs Axe X is Throughput and Axe Y represents density of hitting given Throughput.

That may give you an idea how to compare both results, and that the biggest density is around 3600-3800 tps.

And we are moving to numbers, we can build boxplots:

m <- ggplot(dd,
            aes(x = factor(Tables),y=Throughput,fill=factor(Tables)))
m+geom_boxplot()

That may not be easy to read if you never saw boxplots. There's good reading on this way to represent data. In short - the middle line inside a box is median (line that divides top 50% and bottom 50%), the line that limits the top of a box - 75% quantile (divides 75% bottom and 25% top results), and correspondingly - the line at the bottom of a box - 25% quantile (you should have an idea already what does that mean). You may decide what measurements you want to take to compare the results - median, 75%, etc.

And finally we can combine jitter and boxplot to get:

m <- ggplot(dd,
            aes(x = factor(Tables),y=Throughput,color=factor(Tables)))
m+geom_boxplot()+geom_jitter()

That's it for today.

The full script sysbench-4-16.R with data you can get on benchmarks launchpad

If you want to see more visualizations idea, you may check out Brendan's blog:

• http://dtrace.org/blogs/brendan/2011/12/18/visualizing-device-utilization/
• http://dtrace.org/blogs/brendan/2012/02/06/visualizing-process-snapshots/
• http://dtrace.org/blogs/brendan/2012/02/12/visualizing-process-execution/

And if you're wondering what to do with such unstable results in MySQL, stay tuned. There is a solution.