Archive for March 1st, 2011
This is the second post in a row inspired by a comment on Twitter. The tweet that inspired this is here. The author was expressing her frustration at test hardware that did not reflect the production system.
This approach was recently well illustrated on the BBC Top Gear show as you can see in the excerpt below.
In the case of cars it is rather obvious that substituting a 56bhp, 1 litre 20 year old cheap car for a 512hp, 6.75 litre v8 supercar costing $285,000 is somewhat stupid – hence the comedy value of the clip. In the case of software however we often make similarly stupid decisions. So for example a commodity 2 processor server offers not only different horsepower, but different parallel execution options to a 4 node quad cpu RAC setup. Oracle 18.104.22.168 on Linux is not the same piece of software as Oracle 22.214.171.124 on AIX. A table with 1024 partitions and 100 million rows may well behave differently to an unpartitioned table with 1 million rows.
For straightforward “does this piece of code actually run” testing economizing on test kit and software may be perfectly adequate. For performance testing , it’s the Clarkson approach.