Similar to OLTP, the analytics land scape (online analytical processing, or OLAP) has changed as well. Tra ditionally, relational databases have been used to build data warehouses. This required carefully modeling
a schema in which all the data was preloaded to allow simple analyses using SQL. Again, with the advent of big data, a new generation of systems has appeared, with Hadoop being the exemplary system. Hadoop doesn’t require a fixed schema or preloading the data, nor does it restrict the query language to SQL. Instead, Hadoop lets developers write arbitrary programs by offering a simple programming model (that is, MapReduce), which allows the system to automatically parallelize execution. Even though the advantages sound compelling, the system has been critically received in the research community, and, as mentioned, some argue that Hadoop is a huge step backward.1