| LHM is not slower for iteration (it's faster actually). LHM indeed pays 2 references per node but they are well worth as it has deterministic ordering/iteration and I have witnessed numerous cases with HashMap that show up in production only due to iteration order (esp after rehashing). The code is broken but the cases did not reproduce during testing... Now if the two extra references are an issue, consider that HashMap is quite space inefficient with having a dedicated node per each entry - that's the main price. The (simple) node memory footprint is 36bytes on heaps less than 32GB, i.e. compact pointers. The extra references add another 8 bytes for having an insert or access order. If the goal is getting a compact low memory footprint, HashMap is not fit for purpose. Overall it's the jack of all trades and even got reworked (java8) to support tri-based collision resolution for keys that implement Comparable. Couple years back I wrote CompactHashMap (under CC0) that has an average cost of 10bytes per entry and it's extremely compact for small sizes with having only 2 fields on its right own, so even small/empty maps are tiny. In microbenchmarks (same used in openjdk) it handily (2x) beats java.util.HashMap on "Traverse key or value", get/put have similar performance, and "Search Absent" is worse. The point is: LHM should be the go-to hashmap for java as being node based hashmap is justified (unlike java.util.HashMap) |