Today's internet and cloud services push for lower end-to-end latencies without sacrificing availability or fault-tolerance. However, ensuring fault-tolerance and availability by replicating data across datacenters or even continents comes at a price: higher latency. Many efforts have been made to drive down latency such as by relaxing consistency. This, though successful, is of little interest when ACID transactions are mandatory for most of the internet services. Another direction of research which found success is relaxing fault-tolerance without sacrificing consistency. This method promises sizable reductions in end-to-end latency.
In this work, we analyze the effects of relaxing fault-tolerance on end-to-end latency in the context of a global scale geo-replicated, edge data-management framework. Since different types of transactions may require different fault-tolerance guarantees, we develop a framework in which the fault-tolerance level can be controlled on a per-transaction basis  to minimize end-to-end latency. We shall further introduce a zonal abstraction to encapsulate edge complexity and measure its effects on latency.