Network Attached Storage & Object Oriented Devices

• Why Clusters
• NAS Goal: Enable Scalable Clustering
  • Scalable Connectivity of Storage to Clients
  • Storage Manageability
  • Inherent Security
• Description of NAS
  • Elements of a NAS Configuration
  • Overview of NAS Operation
    • Startup
    • Accessing NAS
  • Data Sharing and Concurrent Update
  • Differences with CMU
• Why Objects

It might seem that there is nothing in the problems of clustering or NAS  that dictates the use of object oriented storage to solve them. In fact object oriented storage does so much to improve the cluster architecture that Seagate feels it is essential to realize the full benefits of the clustered system architecture. Though this is far from doing the subject justice, here are several reasons for this position.

• Objects really make the self-management of storage possible. Without the device having sufficient knowledge of the resident data, it cannot assume the responsibility for managing space. Devices could not participate in any attribute management without the knowledge of what constitutes a meaningful subset of its space or when it is appropriate to take action. More effective management will result from the devices able to participate intelligently.

• The sharing of data can be controlled more intelligently when the device knows what constitutes an entity. If two systems were to share BOD, all the metadata activity would have to be controlled for concurrent access. In an OOD, much of the metadata activity is opaque to the systems, which need only concern themselves with access conflicts to user data. Also space management being done by the device eliminates any contention or confusion that could arise from two systems trying to manage space on the same device at the same time.

• In order for a cluster of systems to truly act as a single computing facility, a single system view of the data is essential. The object-based organization makes this easy.

• Heterogeneous computing is made much easier by an object abstraction. There is essentially no commonality between OS's metadata structures. OODs make it possible to at least have an organization that any OS can interpret.

• While there are a lot of issues revolving around performance in a clustered system architecture, some obvious performance benefits come with the Objects.

  • The metadata never leaves the device, eliminating a certain amount of I/O.

  • The device knows which objects are closed or open and is able to use that information to more effectively cache data.

  • Prefetching can be much more effective as the device knows the layout of the object being read. The device can more effectively determine sequential access patterns.

  • The cache in the device can hold metadata once for multiple systems accessing it.

  • The device can participate in quality of service decisions, such as where to locate data most appropriately. It can only do this if it has responsibility for allocating storage. By comparison, almost no OS can allocate data by zone on a disc drive.