Cloudware Basics - Guide to Cloud Computing for Business and Technology Managers - page 275

are service-enabled, they need to be developed to expand on demand
using scaling-out techniques. The advantage of scale-out systems is
that commodity hardware, such as disk and memory, can be used for
delivering high performance. A scale-out system such as intercon-
nected compute nodes forming a cluster can be more powerful than
a traditional supercomputer, especially with faster interconnect tech-
nologies. Scale-out systems will essentially be distributed systems
with a shared high-performance disk storage used for common data.
Unlike scale-up systems, in order to leverage full power of scale-out
systems, there should be an effort from the programmer to design
applications differently. Many design patterns exist for applications
designed for scale-out systems like MapReduce.
Scale-out solutions have much better performance and price/performance
over scale-up systems. This is because a search application essentially
consists of independent parallel searches, which can easily be deployed on
multiple processors. Scale-out techniques can be employed at application
level as well. For example, a typical Web search service is scalable where two
client query requests can be processed completely as parallel threads. The
challenge in scale-out systems, however, is the complex management of the
infrastructure, especially when the infrastructure caters to dynamic scaling
of resources. Additionally, as noted, applications that do not consist of inde-
pendent computations are difficult to scale out.
13.6.2 Multitenancy
This deals with implementation of multitenancy with fine-grained resource
sharing while ensuring security and isolation between customers and also
allowing customers to customize the database:
1. Ad hoc/custom instances : In this lowest level, each customer has their
own custom version of the software. This represents the situation
currently in most enterprise data centers where there are multiple
instances and versions of the software. It was also typical of the ear-
lier ASP model that represented the first attempt to offer software for
rent over the Internet (see Chapter 11, “Application Service Providers
(ASPs))”. The ASP model was similar to the SaaS model in that ASP
customers (normally businesses), upon logging in to the ASP portal,
would be able to rent the use of a software application like CRM.
However, each customer would typically have their own instance of
the software being supported. This would imply that each customer
would have their own binaries, as well as their own dedicated pro-
cesses for implementation of the application. This makes manage-
ment extremely difficult, since each customer would need their own
management support.