There are a lot of reviews, comparisons and tests regarding databases on the internet. Looking at them more thoroughly, it becomes obvious that these tests are almost always based on a very small data sample. Thus, these reviews are not representative for big production systems.
The database vendors on the other hand publish benchmarks on their web pages that are not usable for production systems either. These numbers should be used with great care and can never serve to compare the speed of two solutions. Load often differs greatly between these tests tests: some use distributed systems, some single node systems. Some tests use a single thread for database access while others use multiple threads.
Thus it quickly became obvious that we had to do our own comparisons for our search engine www.suchen.de in order to better judge the different database solutions.
In relational databases data is subject to a fixed schema: A table usually consists of several columns. Not every column contains an entry. In this case, the databases stores NULL or some other predefined value. This occured quite often in our usage scenarios. In search of a "better" database we had a look a NoSQL data stores, which do not exhibit this problem of "normal" databases.
The products we looked at offered different feature sets, leaving it up to us to choose the best implementation for our use case.

