Pervasive computing introduces data management requirements that must be tackled in a growing variety of lightweight computing devices. Personal folders on chip (e.g., healthcare folders on smartcards), networks of sensors (e.g., pollution sensors) and data hosted by autonomous mobile computers (e.g., tourist information downloaded on a car computer) are different illustrations of the need for evaluating queries confined in hardware constrained computing devices. RAM is the most limiting factor in this context. This paper gives a thorough analysis of the RAM consumption problem and tries to answer three important questions: (1) does a memory lower bound exist whatever be the volume of the queried data? (2) how can a query be optimized without hurting this lower bound? (3) how an incremental growth of RAM impacts the query execution and optimization techniques devised in a lower bound context? Answering these questions paves the way for setting up co-design rules required to calibrate a hardware platform according to given application's requirements as well as to adapt an application to an existing hardware platform. To the best of our knowledge, this work is the first attempt to answer these questions. We illustrate the effectiveness of our answers through a performance evaluation.

Pervasive computing introduces data management requirements that must be tackled in a growing variety of lightweight computing devices. Personal folders on chip, networks of sensors and data hosted by autonomous mobile computers are different illustrations of the need for evaluating queries confined in hardware constrained computing devices. RAM is the most limiting factor in this context. This paper gives a thorough analysis of the RAM consumption problem and makes the following contributions. First, it proposes a query execution model that reaches a lower bound in terms of RAM consumption. Second, it devises a new form of optimization, called iteration filter, that drastically reduces the prohibitive cost incurred by the preceding model, without hurting the RAM lower bound. Third, it analyses how the preceding techniques can benefit from an incremental growth of RAM. This work paves the way for setting up co-design rules helping to calibrate the RAM resource of a hardware platform according to given application's requirements as well as to adapt an application to an existing hardware platform. To the best of our knowledge, this work is the first attempt to devise co-design rules for data centric embedded applications.