Oracle Essentials [Electronic resources] : Oracle Database 10g, 3rd Edition نسخه متنی

14.4 Grid Computing

As with
the Internet (the i in
Oracle8i and Oracle9i),
Oracle Corporation has signaled its intent to embrace the new
paradigm of grid computing (see Figure 14-6) with a
letterthe letter attached to the version number of Oracle
Database 10g. Grid computing is, circa 2003, a
hot buzzword, but the grid computing architecture is the result of
the more powerful aspects of the last two paradigm shifts,
client/server computing and Internet computing.

Figure 14-6. Oracle's vision of grid computing

Grid computing seeks to capitalize
on the most profound economic benefit of the client/server explosion
for ITlow-cost commodity servers. Grid computing is based on
the premise that using multiple commodity machines is less expensive
than using larger SMP machines. Four commodity servers with 4 CPUs
are less expensive than a single SMP machine with 16 serversin
most cases, significantly less expensive.

Grid computing is also the extension of the simplicity provided by
Internet computing. Internet computing relies on a simple user
interface, which requires only a URL to locate a resource. In the
grid, a client simply plugs into computing resources, whose location,
distribution, and topology are completely transparent. The computing
resources that provide the power for the client appear to be seamless
and result in a framework for adding infrastructure resources needed
by applications when they are needed.

14.4.1 Features for Grid Computing

There are several concepts central to the idea of grid computing.
Although none of these concepts are completely new with the grid,
each of them has taken on added importance in this latest
architecture. Before drilling down into further detail, we note that
this book covers grid computing as Oracle has defined it with the
initial Oracle Database 10g release.

The definitions presented here may be somewhat different from other visions or definitions of grid computing you've encountered. Keep in mind that Oracle's vision is also based upon off-the-shelf products rather than upon lengthy custom-built software development efforts.

At the time of this writing, an Oracle-based grid implementation
would consist of database instances running in a RAC configuration
with up to 64 instances running on a common operating system/set of
platforms. The other key components in the configuration are the
following:

Oracle Application Server 10g, the subject of much of this chapter

Oracle Enterprise Manager 10g, discussed here relative to the grid,
but more fully described in Chapter 5

Network-attached storage devices, which will probably use Oracle
Database 10g's Automatic
Storage Management (ASM) capability

Of course, Oracle's database interchange mechanisms
allow the exchange of data with other database grid implementations,
enabling extension beyond a single database. The next sections
describe key software capabilities that enable the building of such
configurations.

14.4.1.1 Provisioning

Years
ago, the computing universe revolved around the sun of the mainframe.
In terms of centralized computing power, the cloud of the grid
occupies the same position as the once mighty mainframe.

The grid is based on consolidating the computing power provided by
multiple servers and clusters. Once consolidated, these resources
must be reallocated to the users and applications that need them.
This allocation, in turn, must be both flexible and transparent for
the grid to deliver on its promised efficiencies.

The mechanism used by the grid for accomplishing this allocation is
termed
provisioning.
A grid computing architecture must be able to match requests and
resources. Because the grid is meant to handle the whole world of
varying requests, provisioning should be easy and able to dynamically
respond to changes in the composition of any particular workload.

One of the requirements for flexible provisioning is the concept of
virtualization,
where resources are delivered in a virtual chunk. The underlying
support for a virtual resource can be changed, in response to demand
or failure, without having to disturb the connection between the
resource and its consumer.

If grid computing is to be achievable without an unacceptable
management burden, provisioning and virtualization are required.

14.4.1.2 Resilience

Consolidation
is a virtue, but
it has a dark side. Does a single computing resource imply a single
point of failure? If so, this one potential problem could outweigh
all the virtues of grid computing.

The grid must be, of necessity, highly
resilient. If one part of the grid fails, the
failure must not endanger the rest of the grid. In order to truly
implement a transparent source of computing power, transparent
failover must be an integral part of the grid.

On the database side of Oracle's grid vision,
Real Application Clusters provide one
part of this type of high availability. If a node in a RAC database
fails, the rest of the nodes just keep on running. The other part of
this equation is supplied by Data Guard, which protects against
disaster scenarios, such as media or site failure.

Oracle Application Server has the ability to cluster its HTTP, Web
Cache, and J2EE servers, as well as the Oracle Internet Directory,
which provides security services to the grid, and Oracle Application
Server Portal, which can deliver a desktop to all users.

14.4.1.3 Integration

From
the outside, the grid looks like one big hunka hunka computing
resource. But there are reasons why there may still be separate
groups of resources, from different locations participating in the
grid to a raft of political and organizational issues. The grid must
be able to integrate these loosely coupled resources together
transparently.

The Oracle database has a few different technologies aimed at
integration. Oracle has supported transparent gateways for quite a
while; these allow a user or application to access data in other
databases with the same API as Oracle. You can access data that lives
outside the Oracle database through the use of external tables.

Oracle
Streams is intended
to be a complete integration solution for data in different Oracle
databases (through replication), as well as between Oracle and other
data sources (through Advanced Queueing).

Web services are another way that separate computing sources can
combine their efforts. The Oracle database allows you to expose
queries (as well as PL/SQL packages) as web services, and can act as
a consumer of other web services. Oracle Application Server also
includes many ways to create, define, manage, and integrate web
services.

Finally, Oracle Application Server includes its set of integration
technologies to bring together all types of heterogeneous services
without too great a management burden.

14.4.1.4 Management

You
may have noticed that we lightly skipped over a huge potential
problem with combining multiple serversthe corresponding
multiplication in management effort. In fact, the increasing
reduction in the price of hardware makes management overhead one of
the largest factors in the overall cost of computing today.

Oracle Database 10g, as described in Chapter 5 and elsewhere, provides a wide range of
features for improved management. The easier it is to manage
individual servers, the lower the overall management costs for the
grid.

Some features described earlier in the book, such as providing
identity management across the enterprise environment, ease the
management burdens that could result from the consolidation of the
grid. Oracle has also expanded the scope and feature set of
Enterprise Manager (EM) in order to help implement and maintain a
grid computing environment. EM changes in Oracle Database 10g
include:

Management of Oracle database and AS

The first big change is the ability to manage both the Oracle
database and Oracle Application Server from within
Enterprise Manager. EM gives you a set
of monitoring and management capabilities for Application Server that
are similar to those available for the Oracle databasefor
example, automatically tracking the time taken by the HTML pages that
are slowest to serve to the client.

Enterprise Configuration Management

Another new area, Enterprise
Configuration Management (ECM), tracks and stores the configuration
details of all monitored servers. You can find out which servers have
a particular configuration option, compare servers against each other
to spot differences, run reports to see which servers have certain
options or vary from an established baseline, or clone a new server
node with a pre-established configuration.

Performance management

One other new area of functionality in EM revolves around the need to
manage user expectations of performance. Performance has always been
a slippery topic: although you may calibrate performance based on
measurements such as CPU cycles or I/O time, your users have their
own, occasionally orthogonal, view of performance as measured by
response time. Enterprise Manager 10g includes a
feature known as Application
Performance Management (APM), which focuses on the end-user view of
performance. The inherent complexity of a grid computing architecture
may be invisible to an end user, but it can still affect the
performance he receives.

APM is implemented by establishing agents, known as
beacons,
to run transactions from various user sites. These transactions can
be defined as any type of user interaction. They are run at regular
intervals, and the results are sent back to the
Enterprise Manager
repository. EM provides graphical historical tracking of these
results, which can help you to quickly see whether a particular
performance problem exists across the grid or whether the source of
the problem is located on the path to a particular location.

You can also set warning levels for transaction performance, which
will cause alerts to be proactively sent if performance degrades to a
specified level.

14.4.2 A Final Thought on Grid Computing

Is grid computing just another marketing buzzword, or is it truly a
new and inevitable destination for the world of IT? We think the
answer is, well, both.

The factors that have created the marketing embrace of grid computing
are certainly reallow-cost machines and simplified end-user
access. And these factors are not only lasting, but will continue to
display considerable appeal to CEOs looking to avoid increases in IT
spending, while still providing more IT services.

Will the IT universe end up as one big grid? It's
not a bad idea, but there are formidable personal and political
barriers to this type of widespread cooperation. Whether the world,
or even one organization, ends up using a grid computing architecture
is, to some extent, beside the point. Everywhere that IT wants to
save money through increased efficiency and consolidation, the
features that benefit grid computing will provide value.