Internal Parallelism of Flash Memory-Based Solid-State Drives

Abstract

A unique merit of a solid-state drive (SSD) is its internal parallelism . In this article, we present a set of comprehensive studies on understanding and exploiting internal parallelism of SSDs. Through extensive experiments and thorough analysis, we show that exploiting internal parallelism of SSDs can not only substantially improve input/output (I/O) performance but also may lead to some surprising side effects and dynamics. For example, we find that with parallel I/Os, SSD performance is no longer highly sensitive to access patterns (random or sequential), but rather to other factors, such as data access interferences and physical data layout. Many of our prior understandings about SSDs also need to be reconsidered. For example, we find that with parallel I/Os, write performance could outperform reads and is largely independent of access patterns, which is opposite to our long-existing common understanding about slow random writes on SSDs. We have also observed a strong interference between concurrent reads and writes as well as the impact of physical data layout to parallel I/O performance. Based on these findings, we present a set of case studies in database management systems, a typical data-intensive application. Our case studies show that exploiting internal parallelism is not only the key to enhancing application performance, and more importantly, it also fundamentally changes the equation for optimizing applications. This calls for a careful reconsideration of various aspects in application and system designs. Furthermore, we give a set of experimental studies on new-generation SSDs and the interaction between internal and external parallelism in an SSD-based Redundant Array of Independent Disks (RAID) storage. With these critical findings, we finally make a set of recommendations to system architects and application designers for effectively exploiting internal parallelism.

Partitioned Real-Time NAND Flash Storage

“…This is called channel striping. Way interleaving and channel striping are the two main methods of parallelization that modern flash controllers support [6].…”

Spatial Indexing on Flash-based Solid State Drives

“…Motivated by the interest of the spatial database and GIS communities on using flashbased Solid State Drives (SSDs), this PhD thesis focused on understanding and improving aspects related to the spatial indexing on SSDs. This is a challenging task since SSDs have intrinsic characteristics that must be take into account to exploit the benefits of these storage devices (AGRAWAL et al, 2008;BOUGANIM;JÓNSSON;BONNET, 2009;KO-UFATY;ZHANG, 2009;KANDEMIR, 2013;HOU;LEE, 2016). For instance, a write requires more time and power consumption than a read.…”

“…In a spatial database context, the replacement of an HDD to an SSD does not mean that spatial applications would inherit all the positive characteristics of SSDs. This is mainly due to the intrinsic characteristics of SSDs (AGRAWAL et al, 2008;BOUGANIM;JÓNSSON;BONNET, 2009;KOUFATY;ZHANG, 2009;KANDEMIR, 2013;HOU;LEE, 2016), which have introduced a new paradigm in data management. A well-known characteristic is the asymmetric cost of reads and writes, where a write requires more time and power consumption than a read.…”

“…The main reason is that SSDs have several improved characteristics than HDDs, such as smaller size, lighter weight, lower power consumption, and faster reads and writes. However, SSDs have intrinsic characteristics that introduce several system implications (AGRAWAL et al, 2008;KOUFATY;ZHANG, 2009;KANDEMIR, 2013;HOU;LEE, 2016), such as the asymmetric costs between reads and writes, the performance interference of interleaved reads and writes, and the read disturbance management. To take into account the intrinsic characteristics of SSDs, flash-aware spatial indices have been proposed in the literature, such as FAST-based indices (SARWAT et al, 2013), the FOR-tree (JIN et al, 2015), and eFIND-based indices (CARNIEL; CIFERRI; CIFERRI, 2017b; CARNIEL; CIFERRI; CIFERRI, 2018).…”

“…But, when it is not possible, a flushing unit should be composed of (i) almost sequential index pages, or (ii) distant index pages (e.g., at least 100 index pages of distance). The former leads to a performance as similar as a sequential write, while the latter leads to a better performance compared to random writes (CHEN; KOUFATY;ZHANG, 2009;KANDEMIR, 2013;HOU;LEE, 2016). Finally, the flushing operation should also take into account Goal 4.…”