ENVS 1200 Lecture 13: ENVS 1200 Lecture 13 Notes
13 May 2018
School
Department
Course
Professor
ENVS 1200 Lecture 13 Notes – Primary Configuration
Introduction
• At least one of the nodes will provide access to the outside world and manage the
cluster.
• The link is used to pass messages and data between nodes.
• It can also be used to repartition the disks as conditions change.
• The workload is divided by partitioning the data between the nodes so that requests
made of each node will be approximately equal.
• This method has the advantage that little communication is required between nodes
because each node is essentially independent.
• The primary difficulty with this configuration is that it is not always possible to plan for
and predict accurately the partitioning.
• As a result, individual nodes may be over-utilized or under-utilized, and the efficiency of
the cluster as a whole suffers.
• The alternative, shared-disk model is shown
• In this model, data may be shared between cluster nodes because of the presence of
disks that are accessible to every node.
• This model offers the advantage of easy dynamic workload balancing and, with careful
design, high availability, and fault tolerance.
• Availability is enhanced on many systems by the use of RAID technology for the shared
disks.
• Although these benefits make shared-disk clusters attractive, there is a cost in software
complexity.
• The software that controls the cluster must be capable of maintaining coordination and
synchronization of the data that is being processed by different nodes
• To prevent corruption of the shared data and to assure accurate retrieval
• For example, suppose that one node attempts to retrieve data that has been modified in
memory, but not yet stored on the shared disk, by another node.
find more resources at oneclass.com
find more resources at oneclass.com
