English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Software-managed heterogeneous memory (HM) provides a promising solution toincrease memory capacity and cost efficiency. However, to release theperformance potential of HM, we face a problem of data management. Given anapplication with various execution phases and each with possibly distinctworking sets, we must move data between memory components of HM to optimizeperformance. The deep neural network (DNN), as a common workload on datacenters, imposes great challenges on data management on HM. This workload oftenemploys a task dataflow execution model, and is featured with a large amount ofsmall data objects and fine-grained operations (tasks). This execution modelimposes challenges on memory profiling and efficient data migration.  We present Sentinel, a runtime system that automatically optimizes datamigration (i.e., data management) on HM to achieve performance similar to thaton the fast memory-only system with a much smaller capacity of fast memory. Toachieve this,Sentinel exploits domain knowledge about deep learning to adopt acustom approach for data management. Sentinel leverages workload repeatabilityto break the dilemma between profiling accuracy and overhead; It enablesprofiling and data migration at the granularity of data objects (not pages), bycontrolling memory allocation. This method bridges the semantic gap betweenoperating system and applications. By associating data objects with the DNNtopology, Sentinel avoids unnecessary data movement and proactively triggersdata movement. Using only 20% of peak memory consumption of DNN models as fastmemory size, Sentinel achieves the same or comparable performance (at most 8%performance difference) to that of the fast memory-only system on common DNNmodels; Sentinel also consistently outperforms a state-of-the-art solution by18%.

Sentinel: Runtime Data Management on Heterogeneous Main MemorySystems for Deep Learning