Research works, whether they aim at building a 'Semantic Web' or a 'Social Semantic Web', consider as a prerequisite that the ideal architecture for managing knowledge would be the Web. Indeed, one can only admire how the CERN internal hypertext scaled out to a world wide level never seen before for this kind of applications. However, current knowledge structures and related algorithms cause new kind of architectural issues. About these issues faced by both communities, we would like to bring out three lessons learnt, three steps in setting up a scalable infrastructure. We will focus on a typical case of knowledge management but with a higher than usual volume of data. Starting with SPARQL, a commonly used Semantic Web technology, we will see the benefits of the REST architecture and the MapReduce design pattern.