Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
A hybrid memory built by SSD and DRAM to support in-memory big data analytics
Chen Z., Lu Y., Xiao N., Liu F. Knowledge and Information Systems41 (2):335-354,2014.Type:Article
Date Reviewed: May 18 2015

Big data analytics use cases are generally characterized by the “3 Vs”: volume (of data), velocity (of arrival), and variety (of data format). For those cases requiring support for high-velocity acquisition and real-time analysis, such as social media sentiment analysis or security incident response, big data architectures typically depend on fast, in-memory data access solutions. Ideally, such access should be dynamic random access memory (DRAM) based, but even as prices for such memory continue to decline, large-scale DRAM systems on the order of many terabytes can be prohibitively expensive. Hybrid DRAM/SSD architectures are often used to reduce costs, although solid-state drives (SSDs) are still an order of magnitude or more slower than DRAM. Thus, the challenge is to optimize the balance of DRAM versus SSDs for both cost and performance. Performance of such hybrid systems is highly dependent on the caching algorithms used to efficiently move data from the SSDs into DRAM. The authors of this paper propose a method for optimizing the DRAM hit ratio based on input/output (I/O) pattern recognition derived from the well-known LZ (Lempel-Ziv) data compression algorithm.

The authors’ results, based on both an adaptive prefetching policy and a novel cache replacement policy, demonstrate that SSD latencies can be masked in a hybrid DRAM/SSD architecture, thus aiding in cost reduction and providing for near-DRAM performance. Their prototype used an Intel-based Linux system with a commercially available SSD, but their results should apply to other types of hybrid memory architectures.

Reviewer:  Harry J. Foxwell Review #: CR143445 (1508-0698)
Bookmark and Share
  Featured Reviewer  
 
Dynamic Memory (DRAM) (B.3.1 ... )
 
 
Information Systems Applications (H.4 )
 
Would you recommend this review?
yes
no

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy