Design and evaluation of an efficient proportional-share disk scheduling algorithm

https://doi.org/10.1016/j.future.2005.09.009Get rights and content

Abstract

Proportional-share algorithms are designed to allocate an available resource, such as a network, processor, or disk, for a set of competing applications in proportion to the resource weight allotted to each. While a myriad of proportional-share algorithms were made for network and processor resources, little research work has been conducted on disk resources, which exhibit non-linear performance characteristics attributed to disk head movements. This paper proposes a new proportional-share disk-scheduling algorithm, which accounts for overhead caused by disk head movements and QoS guarantees in an integrated manner. Performance evaluations via simulations reveal that the proposed algorithm improves I/O throughput by 11–19% with only 1–2% QoS deterioration.

Introduction

The prevalence of streaming services increases the chances for disk resource sharing. As a result, the traffic control on the disk resource called storage Quality of Service (QoS) is gaining in significance in order to satisfy the requirements of different applications [1], [2], [3]. It is known that partitioning such disk resources as bandwidth helps to satisfy the different QoS requirements of various types of applications, such as best-effort applications and real-time applications [3]. Despite its importance, research on storage QoS is still in its infancy, having mainly focused on underlying disk-scheduling algorithms.

A disk resource exhibits different characteristics from other resource types, such as processors and networks, because high overhead is typically involved in processing I/O requests. Few disk scheduling algorithms proportionally share disk resources, such as YFQ [2]and Cello framework [3]. Among them, the YFQ algorithm is based on packet-based fair queuing algorithms in the network, i.e., Weighted-Fair Queuing [4]to choose a subsequent I/O request to be scheduled and Start-time Fair Queuing [5]to maintain a global virtual time. Considering that achieving the ultimate QoS guarantee requires integrated scheduling and management for various resources in the underlying system, such as a processor, network, or disk, the YFQ-like approach with a packet-based fair queuing algorithm is preferable. Unfortunately, proportional-share disk-scheduling algorithms, while preserving a given QoS feature, inevitably suffer from performance degradation in order to improve their disk I/O performance. For example, the YFQ algorithm first selects a batch of I/O requests mainly based on a QoS guarantee, and then attempts to reduce disk head movement overhead by reordering the batched requests. After investigating the operations of this scheduling algorithm, we determined that the effectiveness of disk-overhead reduction is restricted by separating the operation of I/O request selection from the operation of reducing disk overhead. Moreover, while we can achieve a better I/O performance with a larger batch size, the size of the batch cannot be arbitrarily increased in actual systems.

This paper proposes a new proportional-share disk-scheduling algorithm to consider the issues of disk I/O overhead reduction and I/O bandwidth provisioning in an integrated manner, thus overcoming the shortcoming of limited batch sizes found in the YFQ algorithm. Our basic idea derives from the fact that combining the operation of I/O request selection with the operation of reducing disk overhead will increase the odds of reducing disk overhead while maintaining a certain level of QoS guarantee. The remainder of this paper is organized as follows. Section 2gives a description of the proposed algorithm. Section 3provides various simulation results under various synthetic workloads and their analysis. Finally, this paper concludes in Section 4.

Section snippets

The proposed algorithm

Fig. 1 presents a set of components of the proposed algorithm to generate an I/O sequence that not only enhances disk I/O throughput, but also preserves a given level of QoS feature. This section will provide a detailed description of each component. We begin by providing a few notations.

Performance evaluations

This section evaluates the performance of the proposed algorithm by discovering two desirable values of Mdmo and Mqos that not only improve disk I/O performance, but also preserve a given QoS feature with negligible deterioration. We begin by describing the simulation environment for our performance evaluations.

Conclusion and future work

This paper proposed a new proportional-share disk scheduling algorithm that considers both disk characteristics and QoS guarantees in an integrated manner. It consists of a BQS module and a DOR module. The former generates a base I/O sequence based on a typical fair-queuing scheme and the latter inserts extra I/O requests to the base I/O sequence if they meet the two given properties associated with the use of the available overhead in disk head movements and a limited relaxation of QoS

Acknowledgements

This research was supported by the Daegu University Research Grant. The authors also would like to thank the Ministry of Education of Korea for its financial support through its BK21 program. This research was also supported in part by grant no. R01-2003-000-10739-0 from the Basic Research Program of the Korea Science and Engineering Foundation and by HY-SDR IT Research Center.

Young Jin Nam received a BE degree in 1992 from Kyungpook National University, Korea, an MS degree in 1994, and a PhD degree in 2004 from Pohang University of Science and Technology, Korea. He worked with Electronics and Telecommunications Research Institute, Korea from 1994 to 1998, where he engaged in the development of a microkernel-based operating system for a massively parallel computer. He was a visiting researcher at Novell Inc. (USA) in 1995, and at IBM Almaden Research Center in 2001.

References (12)

  • C. Lumb et al.

    Facade: Virtual storage devices with performance guarantees

  • J. Bruno et al.

    Disk scheduling with quality of service guarantees

  • P. Shenoy et al.

    Cello: A disk scheduling framework for next-generation operating systems

  • A. Parekh et al.

    A generalized processor sharing approach to flow control in integrated services networks: The single-node case

    IEEE/ACM Trans. Network.

    (1993)
  • P. Goyal et al.

    Start-time fair queueing: A scheduling algorithm for integrated services packet switching networks

    IEEE Trans. Network.

    (1997)
  • D. Jacobson, J. Wilkes, Disk scheduling algorithms based on rotational position, Technical Report HPL-CSP-91–7rev1,...
There are more references available in the full text version of this article.

Cited by (4)

  • Courier: Multi-dimensional QoS guarantees for the consolidated storage system

    2014, Future Generation Computer Systems
    Citation Excerpt :

    So some non-work-conserving schedulers [16,17] have been proposed to minimize disk seeks through exploiting spatial locality within each application for improving disk performance, while Fahrrad [18] and Argon [19] allocate disk head time to applications for avoiding swing disk head. Moreover, some other publications make [20–23] aim at maintaining high disk I/O efficiency while providing fair throughput distribution. The key idea behind these QoS-aware non-work-conserving schedulers [21–23] is allocating disk head time to applications based on their weight and I/O requests from each application can only be dispatched during its own time slice.

  • A Stackelberg game to derive the limits of energy savings for the allocation of data center resources

    2013, Future Generation Computer Systems
    Citation Excerpt :

    One of the benefits of this mechanism is that its straightforward formulation enables users to understand what will be the response of the system to a specific bid. Current operating systems and virtual machine systems already provide proportional share mechanisms that further simplify implementation from the technological point of view [4–7]. The use of proportional share also enables dynamic pricing based on the load supported by the system by giving an incentive to users to schedule the execution of their applications based on private information; for example, if a service has a high priority, a user will be willing to pay more when the load in the system is high, although it will be reluctant to do so with low-priority tasks.

  • Moneta: A high-performance storage array architecture for next-generation, non-volatile memories

    2010, Proceedings of the Annual International Symposium on Microarchitecture, MICRO

Young Jin Nam received a BE degree in 1992 from Kyungpook National University, Korea, an MS degree in 1994, and a PhD degree in 2004 from Pohang University of Science and Technology, Korea. He worked with Electronics and Telecommunications Research Institute, Korea from 1994 to 1998, where he engaged in the development of a microkernel-based operating system for a massively parallel computer. He was a visiting researcher at Novell Inc. (USA) in 1995, and at IBM Almaden Research Center in 2001. Since 2004, he has been a professor at the School of Computer and Information Technology in Daegu University, Korea. His current research interests include storage architectures with QoS guarantees, object-based storage, and embedded systems.

Chanik Park earned a BE degree in 1983 from Seoul National University, Seoul, Korea, an MS degree in 1985, and a PhD degree in 1988, both from Korea Advanced Institute of Science and Technology, Korea. Since 1989, he has been working for Pohang University of Science and Technology, where he is currently a Full Professor in the department of computer science and engineering. He was a Visiting Scholar with Parallel Systems group in the IBM Thomas J. Watson Research Center in 1991, and a Visiting Professor with Storage Systems group in the IBM Almaden Research Center in 1999. He has served a number of international conferences as a member of Program Committee and he is a member of technical committee in the SIG-Storage Systems in Korea Information Processing Society. His research interests include storage systems, embedded systems, and pervasive computing.

View full text