De-Optimizing Distributed Databases for a Military Use-Case

Abstract

Research in distributed databases has focused on optimizing the performance of reads, writes, latency, isolation levels, and throughput. Over time, we have seen successful advances that push the limits of overall system performance. However, in this field, researchers have mostly explored use-cases stemming from finance, social media, ecommerce, and banking. There are fewer publications which consider military operations and the unique requirements for distributed databases in this context. It follows that optimizing distributed military databases, along the lines of research supporting non-military use-cases, may in fact lead to worse system performance. In these cases, a "deoptimized" database may perform better.

This paper presents a use-case for testing the performance of a distributed military database. This use-case focuses on collecting intelligence and destroying identified targets. Exploring this use-case will allow us to reinforce three main ideas. First, we offer engineers and distributed military database designers concrete performance metrics for their system. Second, we provide a framework for testing and evaluating a distributed military database. The data we gather over multiple wargame simulations will demonstrate isolation levels above READ COMMITTED are associated with a 30-121 percent increase in tactical latency, measured as the average time difference between when a reconnaissance sensor creates data and when a frontline tactical unit accesses that data. Higher isolation levels are also associated with longer times to destroy targets and more targets surviving to the end of the simulation. Lastly, this paper offers new terms, concepts, and variables to consider for the future testing of distributed military databases.