Citation Context

...y optimizing just the layout engine. Brown [4] propose applying task parallelism, which Meyerovich and Bodík [21] implement using Cilk [2] and TBB [23] to weakly scale a CSS subset. Burckhardt et al. =-=[5]-=- likewise apply task parallel Revisions. We achieve strong scaling and on a bigger subset. Attribute grammars Attribute grammars, first introduced by Knuth [16] to define language semantics, are tract...

Citation Context

...upport parallel computation on multicores, taking advantage of the performance benefits of parallelism and incremental computation at the same time by exploiting structural similarities between them [=-=Burckhardt et al. 2011-=-; Acar et al. 2011]. The approach has also been applied to large-scale MapReduce computations in distributed systems [Bhatotia et al. 2011]. 1While these advances chart a viable approach to increment...

Citation Context

...e each task read and update their assigned cells, and have the array merged successfully back into a consistent board state. Such a process may be used to provide high-performance concurrent programs =-=[2]-=- that greatly improve upon the expected performance gained by parallelization alone. However, it is unclear how Revisions handle complex object graphs, as this has not been specifically addressed; nor...

Citation Context

...self-adjusting computation techniques. There has been some research on this problem, but existing solutions work in certain domains and/or use a sub-optimal algorithms for parallel change propagation =-=[5, 6, 18]-=-. 10. Conclusion We present techniques for scaling automatic incrementalization techniques based on self-adjusting computation to large data sets. These techniques enable expressing big-data applicati...

Citation Context

...in one model. Every model element has special operations to shift between versions of it in time. The main difference to our approach is, that they analyze using historical data and a standard imperative language. In this work we only known about the current state, but can use already computed intermediate results from unchanged parts of the model. b) Incremental Evaluation: Incremental evaluation is the “efficient recomputation in response to changes in input data” [29]. It is also called “self-adjustable computing” in the context of imperative programs tracking data and control dependencies [30]. Most works in this field agree on the advantages, such as efficiency, gained at cost of higher memory consumption [31], [32]. However, they propose either language extensions or new languages with the need to explicitly wrap access to data. Using RAG-controlled rewriting, the way of accessing data is not changed, but caching is provided automatically. c) Model Transformation: Model Transformation is an active field of research with many approaches mostly targeting design time. For a better comparison of this work, we classify our approach using a taxonomy [33]: Source and target model in our...

Citation Context

...cratch. Since the work done is often proportional to the change size [6], rather than the total input size, incremental computation can achieve significant performance gains and efficient use of computing resources. To obtain the benefits of incremental computation automatically, researchers in the programming-languages community have proposed compiler- and language-based approaches. While this has been an area of active research for several decades (§7), almost all prior work targets sequential programs—with the notable exceptions of two recent approaches for parallel incremental computation [27, 48]. Importantly, by leveraging a language-based approach, these two prior proposals [27, 48] have demonstrated substantial speedups, thus establishing that the promise of incremental computation can be realized in parallel programs. Specifically, these systems enable efficient and correct incremental updates to the output through the use of new programming languages with special data types (§7), and by requiring a strict fork-join programming model, where threads communicate only at end points (i.e., when forking/joining). These choices reflect a difficult design tradeoff: they provide the compi...