Open Access
Zhang, Jianyong
Graduate Program:
Computer Science and Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
January 13, 2006
Committee Members:
  • Anand Sivasubramaniam, Committee Chair/Co-Chair
  • Mahmut Taylan Kandemir, Committee Member
  • Wang Chien Lee, Committee Member
  • Qian Wang, Committee Member
  • Hubertus Franke, Committee Member
  • storage systems
  • self-managing and self-tuning
  • I/O scheduling
  • quality of service
  • data migration
  • feedback control
The cost of the storage management has been the dominant factor in the Total Cost of Ownership of a computer system. In order to reduce it, a growing trend is to make the storage system self-managing. This thesis investigates two important topics for self-managing storage system design: (1) workload characterization and generation; (2) online QoS enforcement and self-tuning. Understanding workloads is critical for a self-managing storage system design. This thesis presents a synthetic workload generator for TPC-H, an important decision-support commercial workload, by completely characterizing the arrival and access patterns of its queries. The synthesized block-level traces for 22 TPC-H queuries are shown to accurately mimic the behavior of a real trace in terms of response time characteristics for each TPC-H query. Self-tuning is an important aspect of self-management. For a self-tuning storage system, adaptive resource management plays a critical role. One issue is to provide performance virtualization for multiple applications to share a consolidated storage utility in order to meet their Service Level Objecives (SLOs). This thesis presents a 2-level scheduling framework. This framework uses a low-level feedback-driven request scheduler, called AVATAR, that is intended to meet the latency bounds determined by the SLO. The load imposed on AVATAR is regulated by a high-level rate controller, called SARC, to insulate the users from each other. Using extensive I/O traces and a detailed storage simulator, it is demonstrated that this 2-level framework can simultaneously meet latency and throughput requirements imposed by an SLO, without requiring extensive knowledge of the underlying storage system. Business continuance requires non-disruptive data migration, which is another important issue in resource management. In such situation, a migration task utilizes spare bandwidth left by client applications, in order to minimize the impact on them. However, a migration task also has its own requirements. Especially, it is desirable to complete a migration task in a specified time period. This thesis presents an opportunistic data migration scheme with an adaptive rate controller. With extensive experiments, it is shown that the scheme is effective and efficient to meet requirements for migration.