Dynamic data type refinement methodology for systematic performance-energy design exploration of network applications

Abstract

Network applications are becoming increasingly popular in the embedded systems domain requiring high performance, which leads to high energy consumption. In networks is observed that due to their inherent dynamic nature the dynamic memory subsystem is a main contributor to the overall energy consumption and performance. This paper presents a new systematic methodology, generating performance-energy trade-offs by implementing Dynamic Data Types (DDTs), targeting network applications. The proposed methodology consists of: (i) the application-level DDT exploration, (ii) the network-level DDT exploration and (iii) the Pareto-level DDT exploration. The methodology, supported by an automated tool, offers the designer a set of optimal dynamic data type design solutions. The effectiveness of the proposed methodology is tested on four representative real-life case studies. By applying the second step, it is proved that energy savings up to 80% and performance improvement up to 22% (compared to the original implementations of the benchmarks) can be achieved. Additional energy and performance gains can be achieved and a wide range of possible trade-offs among our Pareto-optimal design choices are obtained, by applying the third step. We achieved up to 93% reduction in energy consumption and up to 48% increase in performance. . IntroductionIn the last years, there is a trend toward networks and network applications implemented with the use of embedded consumer devices. The complexity of modern wired and wireless networks combined with the increased interaction with the environment (e.g. in wireless networks) has increased the dynamism of the data access pattern.The dynamism of network applications depends on various factors. One of them is the network traffic. Depending on that, the behavior of functions, as well as the number of times they are executed, differ. The amount of stored data, needed for these functions, varies as well. Thus, dynamically adjustable storage size is a necessity, allowing for freeing of memory when it is no longer needed. Moreover, a static memory allocation at compile time is not an option in the case of most modern network applications, because they do not have a static data access behavior, but rather a dynamic one. This fact means that a static memory allocation at compile time is not efficient at all, because the worst case situation has to be assumed in the beginning and implemented for the whole execution time. Therefore, dynamic memory allocation and management is required (especially in embedded systems). This leads to an increased reliance on Dynamic Data Types (DDTs from now on), the structures, which allow data to be dynamically allocated and deallocated at run-time and provide an easy way for the designer to connect, access and process data [17]. DDTs (with the most common examples being the single and doubly linked lists) can efficiently cope with the variations of runtime needs (e.g. network traffic, user interaction) and the massive amounts of transferred and stored data.Ine...

Component Based Library Implementation of Abstract Data Types for Resource Management Customization of Embedded Systems

“…In [3] the authors present a Dynamic Data Type Refinement methodology (DDTR). Using this methodology the designer can make tradeoffs between performance and energy consumption by selecting different Dynamic Data Type (DDT from now on) combinations from a library (called Matisse) of such implementations.…”

“…Each application can host a number of different containers according to its particular data access and storage pattern in the algorithm. Choosing an improper operator and container implementation for an abstract data type will have significant negative impact on the total amount of memory, energy, bandwidth and cycle budget usage in the embedded system [3].…”

“…Until recently this was tackled by providing customized dynamic data types for each software application individually [3,5,6], which were then added in a library. Nevertheless, these dynamic data types are rather monolithic, thus for each new software application you can either use one of the available dynamic data types or develop a new customization, which has considerable design time cost.…”

Optimization methodology of dynamic data structures based on genetic algorithms for multimedia embedded systems

“…This category includes arrays and lists. Since there are several implementations (called dynamic data types -DDTs) of these containers, as we show in Section 3.1, choosing an improper DDT will have significant negative impact on the dynamic memory subsystem of the embedded system (Bartzas et al, 2006) as in STL (SGI, 2006).…”

“…2008.08.032 experiments to be carried out typically becomes unaffordable even for a small number of DDTs. For instance, in the case of an embedded application including 10 different DDTs that need to be explored for 10 basic relevant implementations of DDTs for multimedia applications (as proposed in Atienza et al (2004), Bartzas et al (2006), Leeman (2003)), the number of experiments (i.e. multiple runs of the application) that need to be performed is 10 10 ; testing all these combinations manually is not feasible.…”