Making lockless synchronization fast: performance implications of memory reclamation