Distributed Systems - Guide to Cloud Computing for Business and Technology Managers - page 68

functional or are malfunctioning. Moreover, the distributed system
could invoke another instance of the failed components along with
the corresponding state of the process (characterized by the program
counter, the register variable contents, and the state of the virtual
memory used by the process) to continue with the process.
• Distributed systems demonstrate invariance or transparency with
reference to characteristics like
• Accessibility, either locally or across networks to the components
• Physical location of the components
• Migration of components from one host to another
• Replication of components, including their states
• Concurrency of components requesting services from shared
components
• Scalability in terms of the actual number of requests or users at
any instance
• Performance in terms of the number and type of available
resources
• Points of failure, be it failure of the component, network, or
response
Performance transparency is achievable through the technique
of load balancing that is based on the replication transparency.
The middleware layer transparently decides on the balancing
decision to select the replica with the least load to provide the
requested service. Furthermore, the performance can also be prevented
from degrading by continuously monitoring the patterns of access to
components and migrating the components appropriately to minimize
remote access.
3.1.1 N-Tier Application Architecture
In the 1980s, the prior monolithic architecture began to be replaced by the
client/server architecture, which split applications into two pieces in an
attempt to leverage new inexpensive desktop machines. Distributing the pro-
cessing loads across many inexpensive clients allowed client/server applica-
tions to scale more linearly than single host/single process applications could,
and the use of off-the-shelf software like relational database management sys-
tems (RDBMS) greatly reduced application development time. While the cli-
ent could handle the user interface and data display tasks of the application,
and the server could handle all the data management tasks, there was no clear
solution for storing the logic corresponding to the business processes being