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

Why Clusters

This section lays the foundation for developing the OOD requirements by positing a system architecture in which network attached storage solves fundamental problems or offers significant functional advantages over traditional architectures.

Figure 1: Mainframe Computing Model

The traditional progression for an end user is to purchase a system and – when additional processing capability is needed – to replace it with a bigger system. This has been the prevailing method for mainframes in particular. At multiple points in this sequence, traumatic discontinuities occur. If the user outgrows the architecture the user started with, the user may have to convert from one operating system to another or even from one vendor's proprietary architecture to another's. Converting entails huge costs for the organization in both dollars and employee time such that these conversions are avoided if at all possible. Another disadvantage with this model is the poor residual value of computer equipment. A system replacement often results in the invested capital being lost when the old system goes out the door. Moreover, larger systems tend to be sold in lower volume than smaller ones, which results in each new system having a higher cost of computing.

Consider the following alternative architecture:

Figure 2: Cluster Computing Model

Replacing a mainframe computer with a cluster of smaller, standards-based servers offers compelling advantages to the user. Since the cluster can start off as a single system, the threshold to entry is lower. These smaller systems are sold in higher volume, making the cost of computing less. With no dependence on proprietary architectures, the availability of equipment from multiple sources lets the user choose the best alternative with each purchase.

There are more advantages. The upgrade cost can be controlled more precisely by adding only the amount of additional resource required. If a different vendor has leapfrogged the original supplier in terms of offering what the user needs at the time of upgrade, the user would be free to choose the best value without concern about migration or conversion to a new architecture. With the right architecture, there might never be a need for conversions from one OS to another, retraining of staff, or developing new procedures – all of which have plagued mainframe customers in the past.

It would be so nice if the cluster computing model were as easy to achieve as it is to draw. There are fundamental architectural roadblocks to be overcome if the cluster model is to realize its potential.

While it is perhaps too simplistic to tie the continued importance of mainframes in the systems market to only two characteristics, there are two significant ones that prevent clusters of open architecture systems from pervasively replacing mainframes. One is the inability of clustered systems to share data in a way that lets the cluster take on the workload that a single mainframe could. The other arises from the extensive experience the mainframe world has in managing storage and data. This experience has evolved into management software that simply has as yet no equivalent in the open systems markets. If clusters of systems are to replace mainframes, these two deficiencies will have to be remedied. This is where NAS comes in.