Databases Reference
In-Depth Information
the system's performance. Next we'll show you how to plot this quickly and see what
happened during the benchmark.
The Importance of Plotting
If you want to achieve world domination, you must plot continually, pun intended.
But seriously, the single easiest and most rewarding thing you can do with your system
performance metrics is plot them in a time series and look at them. You can spot prob-
lems on a chart instantly, when they could be difficult or impossible to see by examining
the raw data. You should resist the temptation to simply look at the averages and other
summary statistics your benchmark tool might print out. Averages are useless, because
they obscure what is really happening. Fortunately, the output from the scripts we've
written so far is custom-made for tools such as
gnuplot
or
R
to plot in the blink of an
eye. We'll demonstrate using
gnuplot
, assuming you saved the data into a file called
QPS-per-5-seconds
:
gnuplot> plot "QPS-per-5-seconds" using 5 w lines title "QPS"
This instructs
gnuplot
to plot the fifth field in the file (the QPS field) with lines and title
it “QPS” on the plot.
Figure 2-2
shows the result.
Figure 2-2. Plotting the benchmark's QPS
Now let's look at an example that will make the value of plotting more obvious. Suppose
your system suffers from so-called “furious flushing” when it gets behind on check-
pointing and blocks all activity until it catches up, causing sharp drops in the through-
put. The 95th percentile and average response times will not show the drops, so the
results will hide the problem. However, a graph will show periodic notches. This is
illustrated in
Figure 2-3
.
