Generic datastructure Datastructure for $k$mers Datastructure for streaming of $k$mers Locality preserving datastructure for streaming of $k$mers. If we make \(B\) buckets, each $k$mer will still need to be mapped to one of \(B\) buckets, requiring \(\lg B\) bits per element still. Interestingly, the first two methods do not depend on \(k\), and involve a lot of cache misses for each lookup, while SSHash becomes better for larger \(k\), since it stores an SPSS (spectrum preserving string set) of fixed size (\(\approx 2\) bits/kmer in the best case, but usually more) and on top of that only stores \(O(\lg n)\) bits of data per minimizer, of which there are fewer for larger \(k\).