Design Space Exploration and Management Policies for Locality-Aware Hybrid Storage Architectures
Open AccessSolid State Devices (SSD) are becoming increasingly important components in the memory hierarchy. They are completely built as semiconductor chips with no need for moving parts. This makes them capable of potentially providing one order of magnitude better response time than rotating disks. The favorable physical characteristics of SSDs coupled with their low cost are what motivated researchers to explore a new design space by incorporating them into the memory hierarchy. However, fundamental architectural decisions have to be made, as to how these devices are integrated into the memory hierarchy. More precisely, should SSDs be used in an augmented hybrid memory in order to remedy the high Dynamic Random-Access Memory (DRAM) costs and power consumption? Should they be used as part of a hybrid disk in order to reduce the magnetic disk hit ratio? Or should they be used to provide an intermediate cache layer between the Random-Access Memory (RAM) and the magnetic disk?SSDs are characterized by their asymmetric I/O properties. For example, NAND SSDs have excellent random read latency; however, they have a relatively slow random write performance. This means that any architecture utilizing these devices should leverage their desirable features while hiding their drawbacks. The primary focus of this work is on how NAND SSDs should be integrated into the memory hierarchy. This work therefore proposes an architectural approach tailored to mitigate the SSD drawbacks while leveraging their performance advantages.The specific problem at hand can be stated as follows. Given a baseline system that consists of a main memory and a magnetic disk, an allocated budget, and I/O intensive workloads; it is required to explore the design space of a dual hybrid cache and storage architecture using several hardware and software configurations. More precisely, the objectives of the proposed work are to:-Explore the design space trade-offs when using SSDs as an extension to main memory.-Explore the design space trade-offs when using SSDs within a hybrid disk architecture.-Explain the impact of combining the two approaches (Dual Hybrid System) on performance and cost.-Develop an intelligent algorithm based on data mining and locality awareness for managing such memory hierarchy, specially the hybrid cache architecture.-Demonstrate a proof-of-concept based on simulation, analysis, and realistic data intensive workloads.-Investigate, where possible, the performance impact of replacing the NAND SSD with other SSD technologies such as Phase Change Memory (PCM)In addition to analyses, the hybrid architectures have been implemented and explored by using and extending the well accepted Carnegie Mellon's storage system simulator (DiskSim). In general, the hybrid cache architecture was the most advantageous as compared to the hybrid disk and the dual hybrid architectures, due to its efficient exploitation of data locality. This hybrid cache architecture coupled with a proposed locality-aware cache management policy has shown that NAND SSD can contribute to up to 35% reduction in the Disk Hit Ratio (DHR) and up to 36% improvement in SSD Hit Ratio (SHR). Additionally, the Response Time Average (RTA) has shown up to 81% reduction compared to 350% increase in terms of I/O Per Seconds (IOPS). Our research findings have also concluded that replacing the NAND SSD with PCM would substantially reduce the hybrid cache response time average by 99% compared to the NAND SSD based hybrid cache architecture. However, using PCM in general remains cost prohibitive.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Aldahlawi_gwu_0075A_11644.pdf | 2018-01-16 | Open Access |
|