1
| 1 |
"Here oft, when Evening sheds her twilight ray, And gilds with fainter beam departing day, With breathless gaze, and cheek with terror pale, The lingering shepherd startles at the tale, How, at deep midnight, by the moon's chill glance, Unearthly forms prolong the viewless dance; While on each whisp'ring breeze that murmurs by, His busied fancy hears the hollow sigh." from "Stonehenge," by Thomas Stokes Salmon, 1823 |
Faculty: Tzi-cker
Chiueh
Current Member(s):
Maohua Lu
Shibiao Lin
Ex-Member(s):
Gang Peng
Lan Huang
Srikant Sharma
Ashish Raniwala
Motivation
To design, build and evaluate a network storage cluster system and research on the issues of storage scalability, storage manageability and quality of service.
Project Description
Persistent information plays a critical role in enterprise and personal computing. Emerging applications such as data warehousing, multimedia content distribution, electronic commerce and medical and satellite image database have substantial storage requirements that are growing at 3X to 5X per year. The emergence of network-attached secure disks (NASD) and storage area networks (SAN) makes storage system easy to scale. Both of them need special hardware such as NASD devices or fiber channel networks. With the elevating gap between the increasing storage management cost and the decreasing hardware cost, storage manageability becomes more and more important. Storage virtualization provides an abstraction of physical devices to policy based storage management software and greatly simplifies storage management. However, currently virtualization is only in capacity dimension and does not care about performance issues. In a storage sharing environment, quality of service must be provided so that various clients can get proper services with quality guarantees. However, none of the previous research ever built such a storage cluster for general applications with QoS in mind.
This project aims to solve the scalability, manageability and quality of service issues in storage cluster by developing a novel clustered storage system called Stonehenge. Stonehenge allows multi-dimensional (bandwidth, capacity, latency, and availability) storage virtualization which greatly simplifies storage management and satisfies demanding storage requirements. The architecture of Stonehenge decouples control path and data path, which makes itself easily scalable. It is built on IDE drives and IP networks and does not need special hardware like fiber channels or NASD devices, which makes it readily deployable within affordable budget. In the process of designing Stonehenge, we designed a measurement-based admission control and a real time disk scheduling algorithm which are the first known algorithms for a virtualized storage cluster system. We also designed a virtual to physical disk mapping algorithm for storage virtualization.
As a first step, we will build a Stonehenge prototype to provide QoS guarantee in terms of basic physical disk metrics. We will evaluate the efficiency of measurement-based admission control and efficiency-aware QoS-guaranteed disk scheduling in this prototype. After this, we will extend the basic Stonehenge system to support virtual disk availability. A generalized real-time scheduling framework for disk array will be developed. Its corresponding admission control will be designed and implemented. Also, a virtual disk recovery strategy is to be developed.
Current Status
Publications
Related Work
Acknowledgement
This project is sponsored by an NSF grant CISE/ACI-0234281 (#0234281).
1. art.com posters prints photos frames, Stonehenge Pictures and Posters.
Last Modified: 7/25/2007 by Gang Peng