THE DECENTRALIZED DATA CENTER

In the Age of Service Oriented Architecture

Jackson He, Enrique Castro-Leon

Digital Enterprise Group, Intel Corporation, 2111 NE 25

Ave, Hillsboro, OR 97124, U.S.A.

Simón Viñals Larruga

Intel Corporation, Plaza Pablo Ruiz Picasso No. 1, Madrid 28020, Spain

Keywords: Data centers, SOA, service oriented architecture.

Abstract: The adoption of SOA in business computing environments is growing due to the promise of significant cost

reduction in the planning, deployment and operation of IT projects. However, the organic transformation

from legacy enterprise applications to SOA applications only has been seen mostly in large enterprises data

center where services are centralized. Patterns of adoption of SOA, in combination with emerging

technologies lead us to believe that the traditional data center owned by a large organization, i.e., the

traditional monolithic data center, will evolve into a more federated form. This new dynamic will affect

large and small business alike.

1 INTRODUCTION

Service-orientation and adoption of SOA (service-

oriented architecture) in business applications have

been discussed in depth in the literature as well as

documenting the SOA transition from conceptual

vision to engineering reality.

However, our research indicates that most IT

managers and end users view SOA as a way for

large enterprises to improve IT efficiency and reduce

costs. A prior paper postulated the evolution of

service orientation around an industry-wide portfolio

of standardized services available to organizations

and businesses large and small. In fact, this

anticipated evolution would ensure that the benefits

of service orientation would be available to a broad

range of organizations, large, small, private and

governmental, and regardless of location or the

maturity of the economy in which they are offered.

The concept of outside-in SOA previously

discussed in a previous paper by the authors

followed by an analysis of patterns of evolution in

the physical infrastructure where these services

reside, namely in data centers. We believe that

patterns in the adoption of SOA will also affect the

way data centers are deployed and sourced.

The decentralization brought by compound service

oriented applications will logically lead to a similar

pattern of decentralization in data centers. This

evolution mirrors the trend toward outsourcing for

development and operations that has occurred in the

past 10 years.

2 INSIDE-OUT: SOA IN LARGE

ENTERPRISES

The increasing adoption of SOA in the industry

brings the promise of significant cost reduction in

the planning, deployment and operation of

Information Technology (IT) projects. Added

regulatory compliance characteristic of this period

has brought additional incentives to explore means

to bring in regulatory relief.

Traditionally, corporate applications have been

deployed in stovepipes, as illustrated in Figure 1,

one application per server or server tier hosting a

complete solution stack. Ironically, this trend was

facilitated by the availability of low-cost commodity

servers fifteen years ago. Under this system

physical servers are procured, a process that takes

anywhere from two weeks to six months. When the

servers become available, they are provisioned with

226

He J., Castro-Leon E. and Viñals Larruga S. (2008).

THE DECENTRALIZED DATA CENTER - In the Age of Service Oriented Architecture.

In Proceedings of the Tenth International Conference on Enterprise Information Systems - SAIC, pages 226-229

DOI: 10.5220/0001682802260229

 SciTePress

an operating system, database software, middleware

and the application. Multiple pipes are actually

needed to support a running business.

Figure 1: Traditional application stovepipes vs. SOA.

With SOA, monolithic applications are broken into

standards-based services.

To cope with the long development and

transition cycle of “internal-only” SOA approach

described above, many enterprise IT departments

pursue a path outlined in Figure 2. While the overall

SOA solution architecture is defined and controlled

by the IT department, some of these services may be

outsourced to external service providers to build and

support (host). This progression starts with small

with non-mission critical services while the IT

internal development teams focus on core,

complicated, mission critical services.

Some companies may be under no such

restrictions, and the core may become so small that

it becomes indistinguishable from the outsourced

services as shown in Figure 3. This evolution for

SOA adoption can be construed as inside-out where

an SOA environment is built out from reusable in

house services out of which a few eventually

become outsourced.

Figure 2: Transition from inside-out to outside-in SOA.

3 DECENTRALIZED SOA

Small business can actually completely sidestep the

inside-out process described above and build their

solutions entirely out of outsourced services. We

call these services servicelets or microservices they

are self-contained hardware and software

components that can be “snapped” together using

Web services technology into full fledged business

applications.

Figure 3 illustrates these ideas. A well defined

business process (e.g. purchase order creation and

processing, bank transactions, etc.) could be

represented by a set of SOA services integrated by

different users depending on their specific business

needs, the process of picking and choosing the right

pieces for their businesses. Mash-ups in the Web 2.0

world today constitute an example.

The model is highly compatible with services

provided by a few large internet portal datacenters

(such as Google, Yahoo, Baidu, Salesforce.com,

etc.) therefore reduce the needs to have large,

company owned, complex centralized data centers

This adoption of SOA will take place under a

different dynamic: instead of reuse from within, or

across organizations in a large enterprise, considered

a necessary condition for critical mass, we will see

the same critical mass, but with reuse now be

happening across whole economic ecosystems. We

call this dynamic an outside-in dynamic for SOA

adoption.

In other words, the outside-in model for SOA

depends on decentralized ecosystem players

providing composite applications that are used to

build more complex SOA-style composite

applications.

Figure 3: For decentralized business computing environment

the core becomes arbitrarily small to the extent that it's no

longer distinguishable.

With such an inside-out SOA model, specialized

SOA services that could be shared by many

businesses could emerge.

The dynamic of this process is that as service

technology matures, services will become and will

be traded as commodities. This will lead to varying

degrees of specialized providers. It simply will be

cheaper to build specific functionality by contracting

out constituent components in the marketplace rather

than build the same functionality wholly in house.

227

THE DECENTRALIZED DATA CENTER - In the Age of Service Oriented Architecture

The fungible, interoperable services mentioned thus

far will have an effect in the way data centers are

deployed and operated. The services mentioned in

the SOA “application” layer are eventually hosted in

a data center. Running these services in a rigid,

monolithic physical infrastructure would to some

extent, negate the benefits of agility.

4 THE PHYSICAL DATA

CENTER INFRASTRUCTURE

What is the physical data center infrastructure

necessary to support such outside-in environment?

The adoption of commodity servers for hosting

enterprise applications started in the early 90s, away

from mainframe and RISC based hosts. At that time

the cost of the hardware was much higher than it is

today.

Commodity servers brought in an attractive

value proposition with the option of additional

capacity in increments of tens of thousands of

dollars, often less than that. Fledgling companies

and rising upstarts found these servers could be

deployed very quickly without long procurement

cycles.

For commodity servers, conventional wisdom

indicates that the norm was a one-application-per-

server provisioning approach. Actually it was worse

than that. Deployments with many servers per

application were not uncommon at all, starting with

the classic 3-tier distributed architecture that splits

storage, application logic and presentation logic

(Web server) in different machines.

Furthermore, an often used mechanism for

improving Web server performance was to split the

front end into several servers, with each server

taking on requests on a round-robin basis.

This approach made sound economic sense at the

time because they brought business agility that was

unattainable with more traditional technologies. It

also provided logical isolation.

Unfortunately, these trends led to companies

deploying thousands of servers. This condition is

affectionately called “server sprawl”. This

development became unsustainable due to space and

energy consumption.

5 THE UTILITY COMPUTING

INFRASTRUCTURE

The new vision for the enterprise architecture is a

multi-layered architecture in which applications

draw on shared services and computing resources.

The utility computing infrastructure is the

foundation for that architecture.

The business engine that is driving the new

vision for the data center is strategic cost reduction

and agility, the same goal as SOA in application

space.

In particular, for SOA to become reality, the data

center infrastructure needs to reflect this new

approach to the point that the infrastructure itself

will become service oriented.

Furthermore, if we take as a given that in the

advanced maturity stages data centers become

policy-oriented and business drives IT, it will be

hard to ignore business considerations both at the

strategic and tactical level.

Under SOA, cross-company barriers may not be

barriers at all. In other words, the trend toward

outsourcing can be expected to accelerate.

Outsourcing will not only encompass applications

under the concept of software-as-a-service (SaaS),

but will also likely include physical plant.

Data centers may become very large, with one

provider taking on one function for several

enterprises, or very small in the case of a specialty

service provider.

A prerequisite under this new environment is that

the architecture, processes and analysis methods

applied will need to be scalable over a much broader

range than what is common today.

6 SOA DEVELOPMENT

In this example we describe an enterprise IT

environment with several key enterprise applications

such as customer relationship management (CRM),

business-to-business (B2B) portals, enterprise re-

source planning (ERP), and others. In a typical IT

environment, these applications are hosted internally

at a centralized data center. However, in the

“outside-in” model, this IT department is

horizontally specialized and does not directly own

any large data centers, just a few servers to integrate

services provided from other service providers. In

fact some of the applications may be renting in a

pay-as-you go SaaS (Software as a Service) model.

This IT department operates by integrating

applications services needed to meet its customers’

needs. Due to the nature of SOA, it is not locked to a

particular services provider. It can easily switch to a

better cost-effective service provider or when the

business needs change.

The trend toward distributed, multi-tier data

center designs has been taking place for the past

ICEIS 2008 - International Conference on Enterprise Information Systems

228

fifteen years. A parallel trend in software has been

the gradual separation of computing engines from

the applications and from the data being operated

upon. Until very recently, an enterprise application

was tightly bound to a physical server and used

direct-attach storage, i.e., data was kept in hard

drives inside the same server boxes. This

arrangement was reasonable from the standpoint that

the data in a hard drive was tightly bound to the

application that created it.

The first boundary to be breached was the one

between compute engines and applications:

advances in software engineering made it possible to

run an application in a multiplicity of platforms.

The emergence of storage area networks (SANs) and

network attached storage (NAS) took the data out of

the boxes. In the past five years the adoption of

XML web services interfaces accelerated this trend

through increasing interoperability and by making

data usable by most any application.

What has not changed in this process is the

prevalence of vertically integrated solution stacks.

Compute engines, applications and data can be

mixed and matched. However, except for very

specific exceptions involving outsourcing, the

common notion is that these three elements do not

cross corporate boundaries. These boundaries will

be breached as well under an outside-in SOA

environment and as part of the technology

maturation process. The reasons will be simple

economics: outsourcing services will lower cost and

yield higher operational efficiencies than an

equivalent in-house solution.

Enterprise application services will be procured

through a hierarchical, multi-layered ecosystem.

Technology maturation makes specialization

possible with opportunities to add value at each

layer.

Figure 4: Creating and sharing typical outside-in services

using a decentralized model.

As illustrated in Figure 4, an independent service

provider goes through the typical service development

process from “model”, “design”, to “implement”.

Once a service is developed, it will drive rigorous

test process to ensure service integrity and quality.

The service then will be transferred to a data center

to publish and deploy. Since services are

independently developed, they could be deployed at

a smaller data center at decentralized locations.

During the service development process, a service

could also invoke services from other service

providers. These offerings can run the whole gamut

of outsourced applications available today, from e-

mail sold by the mailbox (in quantities from one to

several hundred thousand), to services like payroll,

healthcare settlement services, CRM and ERP.

7 CONCLUSIONS

“Outside-in” SOA is still at its early stage of

definition. However, it promises to extend the

benefits of SOA to small and large enterprises alike

through ecosystem supported interoperable services.

SOA in turn will impose requirements of agility on

the physical data center infrastructure with will

change radically the way data centers are deployed

and operated today.

REFERENCES

Tsai, W.T., 2005. Service-Oriented System Engineering:

A New Paradigm. In IEEE International Workshop on

Service-Oriented System Engineering

Tsai, W.T., Xiao, Bingnan, Paul, Raymond A., Chen,

Yinong, 2006. Consumer-Centric Service-Oriented

Architecture: A New Approach, In Proc. of IEEE

International Workshop on Collaborative Computing,

Integration, and Assurance.

He, J., Chang, M., Castro-Leon, E. 2005. Evolution of Intel’s

e-Business Data Center Toward Service-Oriented

Infrastructure. In IEEE International Conference of e-

Business Engineering.

Chang, M,, He, J., Tsai, W.T., Chen, Y. 2006. User-

Centric Service-Oriented Architecture. In IEEE

International Workshop on Service-Oriented System

Engineering.

Castro-Leon, E Chang, M., Y Hahn-Steichen, J, He, J., &

Hobbs, J Yohanan, G., 2006. Service Orchestration of

Intel-Based Platforms Under a Service-Oriented

Infrastructure. In Intel Technology Journal.

Castro-Leon, E He, J, Chang, M., 2007. Scaling Down

SOA to Small Businesses. In IEEE Int’l Conference

on Service Oriented Computing Applications.

229

THE DECENTRALIZED DATA CENTER - In the Age of Service Oriented Architecture