|Name:||(RP09) From Processing-in-Memory to Processing-in-Storage|
|Time:||Tuesday, June 20, 2017
08:35 am - 09:45 am
|Breaks:||07:30 am - 10:00 am Welcome Coffee|
|Presenter:||Roman Kaplan, Technion|
A novel processing-in-storage (PRinS) architecture based on Resistive CAM (ReCAM) is described and proposed. The ReCAM is a massively parallel in-storage accelerator, scalable to tera- and peta-bytes in size. We implemented several algorithms to demonstrate the performance benefits of ReCAM. First is the Smith-Waterman sequence alignment algorithm, which ReCAM can perform in linear time and outperform a cluster of GPUs on the same task. Second is K-means, a key machine learning clustering algorithm in multiple applications, used to group data samples to clusters by similarity. Here we show how ReCAM can perform on a big data dataset and outperform other implementations. Third is online deduplication, a technique for data size reduction in storage systems. We show that a ReCAM-based storage system can provide performance improvements and reduce overhead compared with state-of-the-art storage appliances.
Roman Kaplan, Technion
RP09_Kaplan.pdf (2517 KB)