Resource Allocation Models for Multi-Tiered Storage: Balancing System Efficiency and QoS
| dc.contributor.advisor | Varman, Peter J. | en_US |
| dc.contributor.committeeMember | Cavallaro, Joseph R. | en_US |
| dc.contributor.committeeMember | Zhong, Lin | en_US |
| dc.contributor.committeeMember | Eugene Ng, T.S. | en_US |
| dc.creator | Wang, Hui | en_US |
| dc.date.accessioned | 2016-02-05T15:22:02Z | en_US |
| dc.date.available | 2016-02-05T15:22:02Z | en_US |
| dc.date.created | 2015-12 | en_US |
| dc.date.issued | 2015-11-12 | en_US |
| dc.date.submitted | December 2015 | en_US |
| dc.date.updated | 2016-02-05T15:22:02Z | en_US |
| dc.description.abstract | Multi-tiered storage systems made up of combined Solid State Drives (SSDs) and Hard Disks (HDs) are becoming increasingly popular in shared data centers due to their favorable cost and performance characteristics. Meantime, they are raising new challenges in allocating resources efficiently and providing Quality of Service (QoS) guarantees. Traditional proportional sharing or its generalizations are designed to provide QoS for a single resource type, and lead to poor system utilization when applied to multiple coupled resources. In this thesis we cast the problem of managing multi-tiered storage systems within the broader framework of resource allocation for multiple resources. A fundamental problem that arises when jointly allocating multiple resources is to define fairness policies that provide meaningful QoS guarantees while simultaneously ensuring that system resources are well utilized. We propose a model called Bottleneck-Aware Allocation (BAA), which provides a new definition of fairness for allocation of multiple resources.Based on this notion of per-device bottleneck sets, we design a computationally-effi cient algorithm that maximizes system utilization while meeting re-source capacity constraints and client fairness properties. We show formally that BAA satisfies fairness properties of Envy Freedom and Sharing Incentive. Secondly, we propose a model called Multi-Resource Allocation (MRA), which provides strong quantitative QoS controls including reservations and shares to each client. Reservations specify the minimum throughput (IOPS) that a client must receive, while shares reflect its weight relative to other clients that are bottlenecked on the same device. IOPS based allocation does not differentiate between types of IO requests. This motivates the use of time-based allocation, which considers the variation in request service times. We present Time-Based Bandwidth Allocation (TBBA) to fairly time-multiplex a hybrid storage system while maximizing system throughput. A new allocation model and scheduling are also described. | en_US |
| dc.format.mimetype | application/pdf | en_US |
| dc.identifier.citation | Wang, Hui. "Resource Allocation Models for Multi-Tiered Storage: Balancing System Efficiency and QoS." (2015) Diss., Rice University. <a href="https://hdl.handle.net/1911/88385">https://hdl.handle.net/1911/88385</a>. | en_US |
| dc.identifier.uri | https://hdl.handle.net/1911/88385 | en_US |
| dc.language.iso | eng | en_US |
| dc.rights | Copyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder. | en_US |
| dc.subject | Resource allocation models | en_US |
| dc.subject | QoS | en_US |
| dc.subject | Multi-tiered storage | en_US |
| dc.title | Resource Allocation Models for Multi-Tiered Storage: Balancing System Efficiency and QoS | en_US |
| dc.type | Thesis | en_US |
| dc.type.material | Text | en_US |
| thesis.degree.department | Electrical and Computer Engineering | en_US |
| thesis.degree.discipline | Engineering | en_US |
| thesis.degree.grantor | Rice University | en_US |
| thesis.degree.level | Doctoral | en_US |
| thesis.degree.name | Doctor of Philosophy | en_US |
Files
Original bundle
1 - 1 of 1