Self-awareness and self-adaptation have become primary concerns in large-scale systems as they have become too complex to be managed by human administrators alone, but rather require a new blend of coordination mechanisms between people and software services. This paper presents a roadmap to effective and efficient system adaptation through coupling self- awareness of global-level goals with sustainability constraints. Sustainability of large-scale systems challenges self-adaptation approaches by its intrinsic characters of global and long-lasting effects. We introduce a scientific architecture comprising five levels of awareness: (i) event-awareness, (ii) situation-awareness, (iii) adaptability awareness, (iv) goal-awareness, and (v) future-awareness. Within each level we introduce applicable principles and subsequently outline necessary models, algorithms, and protocols. The approach puts special focus on the interdependencies of human and service elements.