Nitro: a capacity-optimized SSD cache for primary storage
Cheng Li, Philip Shilane, Fred Douglis, Hyong Sop Shim, Stephen Smaldone, Grant Wallace · 2014
For many primary storage customers, storage must bal-ance the requirements for large capacity, high perfor-mance, and low cost. A well studied technique is to place a solid state drive (SSD) cache in front of hard disk drive (HDD) storage, which can achieve much of the perfor-mance benefit of SSDs and the cost per gigabyte effi-ciency of HDDs. To further lower the cost of SSD caches and increase effective capacity, we propose the addition of data reduction techniques. Our cache architecture, called Nitro, has three main contributions: (1) an SSD cache design with adjustable deduplication, compression, and large replacement units, (2) an evaluation of the trade-offs between data re-duction, RAM requirements, SSD writes (reduced up to 53%, which improves lifespan), and storage perfor-mance, and (3) acceleration of two prototype storage sys-tems with an increase in IOPS (up to 120%) and reduc-tion of read response time (up to 55%) compared to an SSD cache without Nitro. Additional benefits of Nitro include improved random read performance, faster snap-shot restore, and reduced writes to SSDs. 1