Loading...

Improving Data Storage Solutions Using Workload Characteristics

Tarihi, Mojtaba | 2021

319 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54810 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Asadi, Hossein; Sarbazi-Azad, Hamid
  7. Abstract:
  8. Responding to the increasing volume and complexity of storage workloads requires continuous design and improvement of storage subsystems. Storage workload behavior such as spatial and temporal locality, request type, and frequency have considerable impact on performance. Hence, performance evaluation and prediction must be performed with respect to workload properties. Moreover, design and implementation of solutions that adapt to workload behavior may further increase the performance and endurance of storage subsystems.One of the key aspects of this thesis is to speed-up the performance evaluation of storage hardware. Three main approaches exist for performance evaluation: simulation, modeling, and measurement from real hardware. Due to the high complexity of storage devices, implementation of accurate simulators is very challenging and thus we focus on accelerated measurement on real hardware and generation of machine learning models to accelerate performance evaluation. For Hard Disk Drives (HDDs), we select a representative subset of the workload to run and achieve as high as 577 speed-up. As I/O workloads modify Solid State Drive (SSD) block layout and can potentially have long term effects on SSD performance, selection of a representative subset is very challenging. Thus, we rely on neural networks and other machine learning models to predict SSD request latency. Another key subject of this thesis is implementation of adaptive algorithms for hybrid SSD caches with the purpose of improving SSD endurance while achieving good performance. For these purpose, it is essential to extract quantitative features from workload behavior so acceleration of storage performance evaluation or implementation of adaptive solutions can be done. We extract features such as temporal and spatial locality, burstiness, request count and request type from storage workloads. Also, due to the lack of suitable SSD workloads, we develop an accurate disk tracing tool and record over 580M SSD workload requests for evaluation and generation of SSD performance models
  9. Keywords:
  10. Hard Disk Drives ; Workload Characteristics ; Solid State Disk Drive ; Storage Workloads ; Storage Subsystems Performance ; Input/Output Intensive Workloads

 Digital Object List

 Bookmark

...see more