Many data sets, such as system logs, are generated from widely distributed locations. Current distributed systems often discard this data because they lack the ability to backhaul it efficiently, or to do anything meaningful with it at the distributed sites. This leads to lost functionality, efficiency, and business opportunities. The problem with traditional backhaul approaches is that they are slow and costly, and require analysts to define the data they are interested in up-front. The authors propose a new architecture that stores data at the edge (i.e., near where it is generated) and supports rich real-time and historical queries on this data, while adjusting data quality to cope with the vagaries of wide-area bandwidth.