Databases Reference
In-Depth Information
Figure 1-2. Readonly benchmarks of several MySQL versions
benchmark with a 2.5 GB dataset, so we configured MySQL with a 4 GB buffer pool.
The benchmark was the standard SysBench read-only workload, with all data in
InnoDB, fully in-memory and CPU-bound. We ran the benchmark for 60 minutes for
each measurement point, measuring throughput every 10 seconds and using 900 sec-
onds of measurements after the server warmed up and stabilized to generate the final
results.
Now, looking at the results, two broad trends are clear. First, MySQL versions that
include the InnoDB plugin perform much better at higher concurrency, which is to say
that they are more scalable. This is to be expected, because we know older versions are
seriously limited at high concurrency. Second, newer MySQL versions are slower than
older versions in single-threaded workloads, which you might not have expected but
is easily explained by noting that this is a very simple read-only workload. Newer server
versions have a more complex SQL grammar, and lots of other features and improve-
ments that enable more complex queries but are simply additional overhead for the
simple queries we're benchmarking here. Older versions of the server are simpler and
thus have an advantage for simple queries.
We wanted to show you a more complex read/write benchmark (such as TPC-C) over
a broader range of concurrencies, but we found it ultimately impossible to do across
such a diversity of server versions. We can say that in general, newer versions of the
server have better and more consistent performance on more complex workloads, es-
pecially at higher concurrency, and with a larger dataset.
