HashMap vs LinkedHashMap performance in iteration over values()

Question

Is there any performance difference between HashMap and LinkedHashMap for traversal through values() function?

Solution 1

I think the LinkedHashMap has to be faster in traversal due to a superior nextEntry implementation in its Iterator

Here is why :

Let us go step by step from the values implementation.
The HashMap implementation of values is this :

public Collection<V> values() {
    Collection<V> vs = values;
    return (vs != null ? vs : (values = new Values()));
}

The LinkedHashMap extends from HashMap and inherits the same implementation.

The difference is in the Iterator implementation for the Values in both.

for HashMap it extends from java.util.HashMap.HashIterator

private final class ValueIterator extends HashIterator<V> {
    public V next() {
        return nextEntry().value;
    }
}

but for LinkedHashMap it extends from java.util.LinkedHashMap.LinkedHashIterator

private class ValueIterator extends LinkedHashIterator<V> {
    public V next() { return nextEntry().value; }
}

so the difference essentially boils down to nextEntry implementation.

For LinkedHashMap it is just calling e.after where e is the Entry , but for HashMap there is some work involved in traversing the Entry[] array to find the next next.

UPDATE : Code for nextEntry() in HashMap

final Entry<K,V> nextEntry() {
    if (modCount != expectedModCount)
        throw new ConcurrentModificationException();
    Entry<K,V> e = next;
    if (e == null)
        throw new NoSuchElementException();

    if ((next = e.next) == null) {
        Entry[] t = table;
        while (index < t.length && (next = t[index++]) == null)
            ;
    }
    current = e;
    return e;
}

The Entry[] is not a contiguous store. (There could be null values in between). If you take a look at the above code, what it does is point next to current and find the next next by iterating over the Entry[] .

But I think this performance gain will come at the cost of insertion. Check out the addEntry method in both classes as an exercise.

Solution 2

I wrote a little profiling program creating 1 million keys (Integer) vs Boolean.TRUE, repeating 100 times. Found the following:

HashMap:-
Create:  3.7sec
Iterate: 1.1sec
Access:  1.5sec
Total:   6.2sec

LinkedHashMap:-
Create:  4.7sec   (30% slower)
Iterate: 0.5sec   (50% faster)
Access:  0.8sec   (50% faster)
Total :  6.0sec

Garbage collection NOT done so pollutes the numbers somewhat, however I think LinkedHashMap has the edge over HashMap and I will be using that in future code.

Solution 3

It almost does not matter. The question is: what do you need. If order of elements is relevant you have to use LinkedHashMap. Otherwise you just do not need it, so use HashMap.

Solution 4

Yes, there will be the same performance difference as you get in all iterations over HashMap versus LinkedHashMap: HashMap will take time proportional to the number of entries plus the size of the hash table, and LinkedHashMap will just take time proportional to the number of entries.

Solution 5

I tried in an UnitTest, iterated values() 10000 times, the milliseconds: 806 vs 902. It is almost the same.

View more solutions

Comments

[exexzian]

could you please elaborate or edit this phrase "but for HashMap there is some work involved in traversing the Entry[] array to find the next next."

[Marcus]

For iterations, it makes sense especially given the above comments, but it's unclear to me how accesses are faster with LinkedHashMap. It makes no sense to me how LinkedHashMap's get(Object key) function which essentially does the same thing as HashMap's get(Object key) function, plus checks an afterNodeAccess() function before returning the result would have a faster performance.

[ACV]

I don't think your test is correct and done properly

[Mark Jeronimus]

Using LinkedHashMap speeds up a certain task from 25 seconds to 8 seconds! I iterate over the entire map for each entry that I insert (to do some value-specific relationship fixing), so O(N^2). (86M nextNode() invocations, hashtable has size 1M and ~13K elements)

[Jeffrey Blattman]

That's an 11% difference. That is significant.