Research

OVERVIEW OBSD MEMS LINKS

The storage sub-systems are always the slowest parts in the modern computer systems. The overall system performances are bounded by the storage systems given moderate amount of I/O operations. Moreover, various applications impose different I/O access patterns to the underlying storage, which makes a universal-optimized file and storage system impractical. Therefore, the general rules of designing a storage system are to capture the typical workload characteristics of the expected working environment and do some special optimizations for such workloads. From my point of view, the former step is more critical. In most cases, the success of a design heavily depends on the understanding of the target environments.

My work has focused on the large distributed object-based storage system. I am particularly interested in the underlying storage optimization for individual storage devices, the file system workload analysis, and the performance scalability problem, especially the metadata server scalability, in the large distributed storage systems. I am also very interested in new storage technologies that are proposed to replace current rotation-based storage devices. Given new access characteristics of those new devices, the design assumptions of current storage systems need to be re-examined.

The OBSD project aims to build a large distributed storage system that supports petabytes storage and provides up to hundreds of gigabytes bandwidth. It is built on the Object-based Storage model, which can provide better scalability, reliability, and security. Two important aspects distinguish the object-based storage model from the traditional storage model: the data path is separated from the controll path and the storage management functions are offloaded to storage devices.

MEMS-based storage is an emerging non-volatile secondary storage technology. With fundamentally different underlying architectures, MEMS-based storage promises seek times 10-20 times faster than hard drives, storage densities 10 times greater, and power consumption an order of magnitude lower.

File System Related Articles

  • SSRC Reading List: for internal users
  • SSRC Paper Archive: for interal users
  • Advanced filesystem implementor's guide

    • Searching Tools

  • Citeseer cs publication search
  • Citeseerbrowser
  • Google scholar search
  • Ask Jeeves
  • Source Code Search Engine

    • Linux Kernel Links

  • Linux Kernel Code Cross-Referencing

    • Programming Links

  • Linux Profiling Tools: Packages that help to detect performance bottlenecks.
  • Linux Debugging Tools: Good or not, you might find something suitable for you.
  • Linux Performance Tools: Tools for detecting Memory Leaks, Profiling, etc.
  • Kernel Module Howto: step by step descriptions on how to port a 2.4 kernel module to the 2.6 kernel.

  • Scientific Softwares: Some useful packages.

    • Software Packages and Traces:

  • File system traces collected from a linux cluster in the Lawrence Livermore Nation Laboratory.
    llnl.tar (2277949440 Bytes, md5sum: 56d765daace01f5c5209d40435652be9)
    llnl.tar.gz (276105363 Bytes, md5sum: a4684c3182d2ce6c4eef662b9ab57762)

  • MEMS-Disk Simulator: based on DiskSim package. This simulator simulates the storage architecture by integrating the MEMS device into the traditional disk device.
    memsvd.tar.gz (505733 bytes, md5sum: 7d00a14a86cd3aae4c1601f48c24801e)

    		•
    Last Modified by Feng Wang
    January 9, 2005