A memory allocator with only 0,006% fixed overhead written from scratch

How can you efficiently allocate tiny (kilo to megabyte sized) objects on petabytes of storage? Write a memory allocator from scratch which uses less than two bits per 4KB. This talk focusses on an interesting aspect of the SLASH/ storage engines for Varnish-Cache which uses a simple yet super efficient buddy memory allocator, a very well known algorithm from the 1960s. The primary novelty of this implementation is that it uses a fixed amount of metadata at about two bits per minimum page size. It also supports prioritized fair waiting allocations and an interactive live view.