Draft 4 User Mode Memory Page Allocation A Silver Bullet For Memory Allocation?

The literature of the past decade has discussed a number of con-trasting ways in which to improve general allocation performance. The Lea allocator, dlmalloc [1], aims for a reusable simplicity of implementation in order to improve third-party customisability and performance, whereas other allocators have a much more complex implementation which makes use of per-processor heaps, lock-free, cache-line locality and transactional techniques [2–4]. Many still believe strongly in the use of custom application-specific allocators despite that research has shown many of these implementations to be sub-par [5], whereas others believe that the enhanced type infor-mation and metadata available to source compilers allow superior allocators to be implemented at the compilation stage [6–8]. More recently there appears to be an increasingly more holistic view of how the allocator’s implementation strategy non-linearl