1. mvn clean install

2. cd target

3. java -jar anagram.jar txt-file-name

Performance:

1. Filtering input dataset. My solution works only with unique set. English language contains only 1_000_000 words. And if try to upload 100 billion, unique words will be only 1_000_000 or less))
2. Working with set in parallelStream();
3. For defining as anagrams, program compares two sorted arrays of chars.
4. Storing anagrams in map. The map which contains key – is a sorted word, the value - is a collection of anagrams. For map interface I used ConcurrentHashMap, and for collection Collections.synchronizedList(new ArrayList<>()). Program works in multithreading, that’s why I used concurrent implementation.

Scalability:

1. You can run my code in multi-nodes environment and feed the node with some parts of data;
2. My class Anagram works with unique words set and returns list of anagram collections;

Maintainability: I think my code is easy to maintain, expand and support. One class stands for one responsibility.