Hello Scott, while digging deeper into this problem I noticed that the data that we store for each node id is extremely redundant. I think it should be possible to build a directory of all possible combinations of values, so that we just have to store one int value which addresses the entry in the directory. Since we are not even interested in the order of the stored vales (the writer ids), this should be a very simple dictionary. I am just not sure if this dictionary should be build in the add method "on the fly" like in zip algorithmns, or if we can compute some tables like lexx+yacc first. Any comments? Gerd -- View this message in context: http://gis.638310.n2.nabble.com/splitter-memory-usage-tp6935688p6946974.html Sent from the Mkgmap Development mailing list archive at Nabble.com.