TL-Tree: flash-optimized storage for time-series sensing data on sensor platforms

  • Authors:
  • Huan Li;Dong Liang;Lihui Xie;Gong Zhang;Krithi Ramamritham

  • Affiliations:
  • Beihang University, China;Beihang University, China;Beihang University, China;Carnegie Mellon University;IIT Bombay, India

  • Venue:
  • Proceedings of the 27th Annual ACM Symposium on Applied Computing
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

Extending NAND flash to off-the-shelf sensor platforms has great potential for improving in-network processing for sensor networks. However, due to the specific read/write characteristics of NAND, the strict resource constraints of sensor devices, e.g., main memory and energy, and the time-series sensing property of sensor applications, designing an efficient resource-aware flash storage system on sensor platforms is a challenging work. In this paper, we propose Time-Log Tree (TL-Tree), a novel indexing structure that is designed to consider time-series as a primary characteristic for optimizing both memory and energy constraints. TL-Tree is an unbalanced tree tailored to temporal sensor data. It realizes the flash utilization bound problem and builds a cascaded structure wherein only the root tree and the most recently updated subtree are stored in the main memory. We prove that TL-Tree minimizes the memory cost and also maximizes the effective usage of flash capacity. Compared to other schemes, simulation results show that the TL-Tree achieves much better access and energy savings for MicaZ platforms. We developed a hardware board that includes a raw 128MB NAND flash chip on MicaZ mote, and also implemented a flash driver and the TL-Tree to demonstrate the practical use of this idea.