Abstract: When selecting a database, users are forced to trade off high throughput for scalable storage capacity. Specifically, a choice is made between performant single machine databases (DBs) or distributed DBs, whose throughput is several orders of magnitude worse. We work towards a novel distributed DB architecture that aims to scale while also sustaining the throughput of single machine systems. We exploit the skew common in real-world workloads and co-locate the most popular keys on a single “hot” node. Although conventional load balancing strategies maximize throughput by distributing hot keys across nodes, we instead choose to create a central point to which targeted, aggressive optimizations can now be applied. In particular, we plan to replace the hot node with a performant in-memory database and layer on a new commit protocol that utilizes its presence to increase throughput while preserving strict serializability by coordinating two distinct databases. In this talk, I will discuss our current progress towards implementing this architecture and our initial findings. We experimentally validate the viability of this approach, showing potential gains to be at least two orders of magnitude greater than that of existing strictly serializable distributed databases.