The document presents a modified pure radix sort algorithm for sorting large heterogeneous data sets. It discusses problems with traditional radix sort algorithms and previous work optimizing radix sort. The proposed algorithm divides the data into numeric and string clusters. It then distributes the numeric data into subsets of equal length which are sorted in parallel using an approach that bypasses certain digits in each pass. String data is sorted by assigning numbers to identical strings. The algorithm is tested on two machines and shows improved performance over traditional radix sort and quicksort, providing sorting times 10-20% faster for large heterogeneous datasets.