The Internet Encyclopedia (Volume 3)

P1: IML/FFX P2: IML/FFX QC: IML/FFX T1: IML

Web ̇QOS WL040/Bidgoli-Vol III-Ch-58 July 16, 2003 9:36 Char Count= 0

Web Quality of ServiceWeb Quality of Service

Tarek Abdelzaher,University of Virginia

Introduction to Web QoS 711

Web Architecture and QoS 711

The Challenge of QoS Guarantees 711

Current Web Architecture 712

The HTTP Protocol 712

Caching and Content Distribution 712

Performance Guarantees in Web Servers 713

Performance Isolation 713

Service Differentiation 715

Challenges 718

QoS Adaptation 719

Performance Considerations in Web Proxy

Servers 721

Conclusion and Future Trends 721

Glossary 722

Cross References 722

References 722

INTRODUCTION TO WEB QOS

The Web has become the preferred interface for a grow-

ing number of distributed applications with various de-

mands for reliability, availability, security, privacy, timeli-

ness, and network bandwidth. These properties are often

called Webquality of service(QoS) dimensions. In this

chapter, we first review the main components of the Web

architecture and describe the protocols that govern their

interaction. We then discuss how this architecture and

these protocols are affected by the different considera-

tions of achieving quality of service that emerge in the

context of contemporary Web applications.

WEB ARCHITECTURE AND QOS

The first question this chapter needs to address is why

Web QoS emerged as a new challenge area, and what

makes this challenge important. Today, the World Wide

Web is by far the largest source of Internet flows. The

great majority of Internet connections use HTTP, which

is the protocol that governs Web access. Improving

its performance, therefore, has dramatic global effects.

Efforts to improve performance come in two flavors. First,

infrastructure improvements are pursued, such as real-

izing higher bandwidth, faster servers, and better last-

mile technologies. This is largely a hardware problem that

motivates development of faster processors, memory, I/O

devices, and communication fabrics. Concurrently, a sub-

stantial amount of research is done to make Web perfor-

mance morepredictable. Performance is said to be pre-

dictable when its quality can beguaranteedin advance.

Many societal and commercial forces contribute to this

need. In particular, the commercialization of the Internet

and the pricing of many Internet services play a signifi-

cant role in elevating the idea of performance guarantees

from a value-added option to a primary concern driven by

contractual obligations.

In traditional commercial products, consumers have

grown to take quality guarantees for granted. Vendors

have contractual obligations to accept returns of defective

products or products that do not perform as advertised.

Similarly, paying consumers of Internet-based services

will soon expect a performance guarantee or a money-

back statement. Much as with other services, it will be

important that clients and service providers be able to ne-

gotiate mutually acceptable quality levels in the service

contract.

Quality of service can be directly tied to revenue. This

relation is manifested today in several domains. For ex-

ample, ISPs often sign mutual service level agreements

(SLAs), which among other things describe the perfor-

mance that the traffic of one ISP should receive in the net-

work of the other and the corresponding fee. Closer to the

end user, online trading services sometimes tie their com-

mission fees to performance in executing the trades. The

fee is waived for trades that are delayed by more than a

certain amount of time. In view of this emphasis on perfor-

mance as a contractual obligation, a significant amount

of research has been spent on architectures for achiev-

ing quality of service guarantees on the Web. In the rest

of this chapter, mechanisms and policies for QoS provi-

sioning, QoS negotiation, and utility optimization will be

discussed.

THE CHALLENGE OF QOS

GUARANTEES

QoS attributes in general can be classified into two cat-

egories, temporal and nontemporal. Temporal attributes

are those related to the passage of time. Their relation to

time can be either direct (such as response time and queu-

ing delay) or inverse (such as throughput). Examples of

nontemporal attributes include security and fault toler-

ance. Guarantees on nontemporal attributes are, in some

sense, an intrinsic function of the algorithms that provide

them. For example, an encryption algorithm generally

has well-defined properties such as the complexity of the

encryption code. These properties directly represent the

security guarantee given to the user. Temporal properties,

on the other hand, are a function not only of their provid-

ing algorithm, but also of external environmental factors

such as resource availability and client load. A schedul-

ing algorithm, for example, cannot provide a guaran-

tee on maximum queuing delay without prior knowledge

of hardware speed and the worst-case client arrival pat-

tern. Unfortunately, this information is often unavailable

a priori, thereby complicating the performance guarantee

problem.

711