Graph-structured datasets such as social networks or knowledge graphs are ubiquitous across all scientific domains. Many current and forthcoming datasets are huge, representing billions to trillions of interactions and connections. In addition to being large, these datasets tend to be irregular and complex, which can make performing even basic analytics on them computationally challenging. Further, modern high-performance computational platforms are themselves large and complex. It is an incredible challenge to concurrently map efficient and scalable graph analytics to both these datasets and these computational platforms. This talk will discuss work towards addressing these challenges at the most basic level: via the 'layout' of the data. Here, layout refers to the way that these datasets are distributed, partitioned, and ordered in-memory on a large-scale parallel architecture. Layout optimization can enable or accelerate computations at massive scales by minimizing communication, optimizing for concurrency, and taking advantage of the compute hierarchy on current and forthcoming large-scale systems.