If we need T log2 T comparisons (where T is the number of termID–docID pairs) and
two disk seeks for each comparison, how much time would index construction for
Reuters-RCV1 take if we used disk instead of memory for storage and an unoptimized
sorting algorithm (i.e., not an external sorting algorithm)? Use the system
parameters in Table 4.1.