Matthew Burke

I believe the crux of systems research is identifying abstractions for systems that are intuitive for application developers and allow underlying implementations to meet application performance requirements. Developers prefer strong abstractions that make it easier to program correct applications, such as those that mask concurrency or failures. However, strong abstractions impose high costs. Consequently, developers of applications with high performance requirements often either resort to weaker abstractions that increase code complexity or limit the scope of their application to reduce performance demands (e.g., by partitioning the userbase).

I aim to address this crux across the “distributed stack” from globally distributed to intra-datacenter to cross-cloud systems. As online services become more ubiquitous, they must reliably and securely process user data using distributed systems. At the same time, these services are supporting increasingly larger userbases with expectations of interactivity when sharing data. Existing abstractions for distributed systems require implementations that fundamentally cannot satisfy these evolving robustness and performance demands. Furthermore, new hardware and software building blocks change these fundamental costs and create opportunities for co-designing systems with their underlying layers to meet performance demands. My research focuses on new system designs that provide alternatives to prevailing abstractions with similar useability and reduced implementation costs.