Poster 121: HFlush: Realtime Flushing for Modern Storage Environments

Authors: Jaime Cernuda (Illinois Institute of Technology), Hugo Trivino (Illinois Institute of Technology), Hariharan Devarajan (Illinois Institute of Technology), Anthony Kougkas (Illinois Institute of Technology), Xian-He Sun (Illinois Institute of Technology)

Abstract: Due to the unparalleled magnitude of data movement in extreme scale computing, I/O has become a central challenge. Modern storage environments have proposed the use of multiple layers between applications and the PFS. Nonetheless, the difference in capacities and speeds between storage layers makes it extremely challenging to evict data from upper layers to lower layers efficiently. However, current solutions are executed in batches, compromising latency; are also push-based implementations, compromising resource utilization. Hence, we propose HFlush, a continuous data eviction mechanism built on a streaming architecture that is pull-based and in which each component is decoupled and executed in parallel. Initial results have shown RFlush to obtain a 7X latency reduction and a 2X bandwidth improvement over a baseline batch-based system. Therefore, RFlush is a promising solution to the growing challenges of extreme scale data generation and eviction shortcomings when archiving data across multiple tiers of storage.

Best Poster Finalist (BP): no

Poster: PDF
Poster summary: PDF


Back to Poster Archive Listing