Abstract
dc:descriptionNon-Volatile Memories (NVMs) have revolutionize data storage replacing traditional magnetic HDDs in both consumer and enterprise computer systems. Flash based Solid State Drives (SSDs) have become the storage medium of choice for many applications thanks to their high throughput/low latency, shock resistance and low power consumption. Lately, the advent of 3DXPoint with even better properties introduced a new breakthrough for storage systems. Data indexing has been significantly influenced by these advances. Indexes are special purpose data structures, designed to provide fast access to large data collections. They have been extensively studied in DBMSes assuming HDDs as the underlying storage. However, treating SSDs as simply another category of block devices ignoring their inherent idiosyncrasies (e.g. erase-before-write, wear-out, asymmetric read/write speed) and their assets (e.g. their ability to process more than one I/O requests in parallel) may lead to poor performance.In this dissertation we survey the most important flash-aware indexes and we present new indexing techniques for the NVM based storage. Briefly, in the first part of our research we study flash aware spatial indexes, while in the second we follow a different roadmap exploiting both flash and 3DXPoint technologies. Thus, we introduce GFFM and LB-Grid two variants of Grid File for flash SSDs. GFFM utilizes a buffering strategy that exploits batch writes, while LB-Grid uses logging to reduce small random writes at the buckets' level. We present flash efficient algorithms for range, kNN and group point queries for both LB-Grid and GFFM. We also discuss our contribution in the development of flash efficient bulk-loading, bulk-insertion and querying algorithms for the XBR+-tree. In the sequel, we examine hybrid storage configurations that comprise flash and 3DXPoint SSDs. We propose the H-Grid, a spatial index structure that utilizes flash SSDs as mass storage tier and 3DXPoint ones as performance tier. We examine R-tree efficiency on a 3DXPoint SSD as well, and we present the sHR-tree an initial, yet illuminating effort to develop a hybrid index based on R-tree. Finally, we assay the latest advances in the SSDs' architecture, the programming models and upcoming NVM technologies and we discuss the future trends and new lines of research related to this field.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Fevgas, Athanasios
- Φεύγας, Αθανάσιος
Subjects
dc:subject × 6Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/46099