Efficient Real-Time, Policy-Based Storage Performance Management for IaaS

Technology #15546

Questions about this technology? Ask a Technology Manager

Download Printable PDF

Categories
Researchers
Tao Li
Ruijin Zhou
Managed By
Richard Croley
Assistant Director 352-392-8929
Patent Protection
US Patent 9,772,794

Hybrid Software Architecture Combines Copy-On-Write/Copy-On-Read Snapshot with Virtual Disk Migration

This storage management architecture enables more efficient Infrastructure as a Service (IaaS) offerings for cloud computing and big data. The global IaaS market’s compound annual growth rate should be 42.91 percent between 2015 and 2019, about a value of $34 billion. Because private data centers have steady input-output (I/O) behavior, they can predict storage latency from usage patterns. They can make workload balancing decisions and storage migrations when they know the system is not in use. In contrast, IaaS providers give 24/7 access. Since data traffic can change instantaneously, it is difficult for IaaS to detect and respond to workload surges. Even if a system keeps a strict scheduling policy, I/O behaviors are unpredictable. Real-time policy-based storage performance management is impossible. Many virtual storage disks can store several Terabytes, so moving them to another location could take hours. Researchers at the University of Florida have developed Light Storage Resource Management (LightSRM), a hybrid software architecture combining copy-on-read/copy-on-write data snapshots with virtual disk migration. IaaS providers can use LightSRM in real-time to balance workloads and tune performance.

Application

Efficient, real-time, policy-based storage performance management for IaaS

Advantages

  • Reduces storage latency, optimizing IaaS efficiency and performance
  • Reduces storage management overhead, improving the accuracy of policy execution
  • Compatible with available IaaS equipment, enhancing versatility
  • Separates performance critical data from virtual disk, enabling fast performance management

Technology

The cloud management architecture includes a virtualized server, multiple storage pools, and a centralized data center management server. The virtualized sever includes a base disk and a virtualized disk with copy-on-write and copy-on-read snapshot technology, which it uses like a cache. When a program wants to make a write operation, the system saves a data chunk snapshot to a new storage pool. When a program wants to make a read operation, the system reads the desired data chunk from the newest location. If the newest location is the base drive, the system fetches the data from the base drive once and copies it to the snapshot file location. To avoid fragmentation and the indexing performance cost, the system will merge old snapshots into the base drive whenever it makes a new snapshot. Since the I/O size is usually less than the virtual disk size, this snapshot process will usually be much faster than migrating an entire virtual disk. If the virtual disk is smaller, then migration is faster. The system dynamically picks the method with the least performance impact using a self-adjusting sliding window tracking mechanism. In experiments, hybrid LightSRM provided 91 percent lower storage latency, 93 percent better balanced storage environments, and 91 percent less disk I/O traffic than standard virtual disk migration.