REST and Web Services: In Theory [612622]

Chapter 2
REST and Web Services: In Theory
a n di nP r a c t i c e
Paul Adamczyk, Patrick H. Smith, Ralph E. Johnson, and Munawar Hafiz
Abstract There are two competing architectur al styles employed for building Web
services: RESTful services and services based on the WS– /ETXstandards (also known
as “SOAP Web services”). These two style s have separate follower bases, but many
differences between them are ideological rath er than factual. In order to promote the
healthy growth of Web services research and practice, it is important to distinguish
arguments for implementation practices ove r abstract concepts represented by these
styles, carefully evaluating the respective advantages of RESTful and WS– /ETXWeb
services. Understanding these distinctions is especially critical for the development
of enterprise systems, because in this domain, tool vendors have preferred WS– /ETX
services to the neglect of RESTful solutions. This chapter evaluates some of the key
questions regarding the real and perceived distinctions between these two styles of
Web services. It analyzes how the current tools for building RESTful Web services
embody the principles of REST. Finally, it presents select open research questions
to further the growth of RESTful Web services.
Introduction
Since its inception, the Web has been an open frontier of exploration in software
and network system design. New ideas were tried and tested first, but organized
and standardized later, once they proved their utility. For example, HTTP, the
transport protocol of the Web, had been in use for more than half a decade before
its state of practice was written down as HTTP/1.0 ( Berners-Lee et al. 1996 )
in May 1996. But the standardization process continued until 1999, when the
final revision of HTTP/1.1 ( Fielding et al. 1999 ) standard was completed. The
architectural principles behind HTTP and other Web standards were described by
P. Adamczyk (/envelopeback)
Booz Allen Hamilton Inc.
e-mail: [anonimizat]
E. Wilde and C. Pautasso (eds.), REST: From Research to Practice ,
DOI 10.1007/978-1-4419-8303-9 2, © Springer Science+Business Media, LLC 201135

36 P. Adamczyk et al.
Fielding (2000 ), thus completing the process. HTML has followed a similar path.
It started out with a simple set of tags for structuring text and graphics on Web pages.As the number of content types [new multim edia formats, more sophisticated ways
of displaying text, interactive Web pages ( Garrett 2005 )] grew, the HTML tags were
pressed into service of displaying them in various non-standard ways. After nearlytwo decades of this growth, new multimedia HTML tags are finally going to beadded and standardized by W3C in HTML5, which is expected to be completed in2012 ( Hickson 2010 ).
A similar sequence of events – simple beginnings leading to an unruly explosion
followed by some type of organization – can be observed in the realm of Webservices. The first Web services were built for passing remote procedure calls(RPCs) over the Web. The idea took off quickly and resulted in a large collectionof standards (beginning with SOAP and WSDL). Surprisingly, these standards weredefined with little consideration for the contemporary practice; sometimes beforethere were any implementations to standardize. The end result of this prematurestandardization was confusion, rather th an order that standards usually bring. In
response, an alternative style of Web ser vices, built according to the rules of the
Web, began to appear. These (so-called RESTful) Web services are maturing, or,more precisely: people are re-learning to use the tried-and-true standards of the Weband applying them when building Web serv ices. As the two styles of Web services
are used side-by-side, one hopes that they will begin to have positive effects on oneanother. Currently, the interactions an d comparisons begin to reach a constructive
stage, so this is a good time to stop and reflect on the current state of affairs.
In particular, this chapter focuses on the interpretation of the widely used term,
REST. Roy Fielding coined the term and codified it under four principles. Inpractice, people are implementing it in man y ways, each harborin g certain implicit
conventions of the developers. Following the path of practice dictating the standards,we raise questions about the previously accepted views about REST and Webservices, and identify the challenges raised by the current state of practice.
Having a standard meaning of RESTfulness would engage the enterprise commu-
nity. REST has been an important part of “r enegade” Web services, appealing more
to independent, small-scale and “hip” dev elopers. With concerted research effort,
it would fulfill the stricter requirements of enterprise Web services; conversely, theenterprise services would benefit from its simplicity.
We begin by summarizing the theory behind RESTful Web services, and draw
a comparison with WS–
/ETXservices. Next, we look into the usage patterns of Web
services in practice: both RESTful services and WS– /ETXservices. Then, we discuss
some of the problems facing the existing RESTful services, how these problemsmake it harder to apply RESTful services to large enterprise systems, and how toolsfor implementing them help to alleviate these problems. We conclude by surveyingsome of the outstanding research problems of RESTful Web services.
Conventions used in this chapter. We consider two dominant styles of Web ser-
vices: RESTful and WS–
/ETX. The term Representational State Transfer (REST) was
coined by Roy Fielding to identify an arch itectural style based on a set of principles

2 REST and Web Services: In Theory and in Practice 37
for designing network-based software architectures ( Fielding 2000 ). Subsequently,
the term was extended to describe a styl e of building Web services based on the
principles of REST. We use the term RESTful to refer to the Web services built
according to this architectural styl e (or parts of it). We use term WS– /ETXto refer to
services based on SOAP, WSDL and other WS– /ETXstandards (e.g. WS-Addressing,
WS-Security), which were defined s pecifically for Web services.
Web Services in Theory
Although this task was undertaken many times before, presenting a fair comparison
of WS– /ETXand RESTful Web services remains a daunting task. In this section, we
will describe their guiding principles and summarize two studies that compare thesearchitectural styles.
Principles
Roy Fielding documented REST based on the principles that emerged as the Webevolved ( Fielding 2000 ). He noticed that Web servers, clients, and intermediaries
shared some principles that gave them extensibility to work on the large-scaleof the Internet. He identified four principles of REST (which he called con-straints) ( Fielding 2000 ):
1. Identification of resources.
2. Manipulation of resources through representations.3. Self-descriptive messages.4. Hypermedia as the engine of application state (abbreviated HATEOAS).
These principles describe the architect ure of systems and interactions that make
up the Web. The building blocks of the Web are called resources . A resource
is anything that can be named as a target of hypertext (e.g., a file, a script, acollection of resources). In response to a request for a resource, the client receives arepresentation of that resource, which may have a different format than the resource
owned by the server. Resources are manipulated via messages that have standard
meanings; on the Web, these messages are the HTTP methods. The fourth principlemeans that the state of any client–server interaction is kept in the hypermedia they
exchange, i.e., links, or URIs. Any state information is passed between the client andthe server in each message, thus keeping them both stateless. It’s easy to check anydesign against such a simple description. Any discrepancies will be easy to identify.However, this simplicity is deceptive – if one tries to s implify it even more, the
entire design suffers. We will discuss concrete examples of oversimplifying RESTin some Web services in “REST Concepts in Practice”.

38 P. Adamczyk et al.
WS–/ETXservices do not have a single metaphor. Web Services Architecture
document ( W3C Working Group Note 2011 ) from W3C describes four architectural
models of WS– /ETX, but does not explain how they relate. One of the models is the
Resource Oriented Model (which would imply REST), but as their definition ofWeb services suggests, the systems they consider are limited to various standards:SOAP, WSDL, and others. New capabilities are added to WS–
/ETXin the form of
new standards. There is no overarching description of the relationship betweenWS–
/ETXstandards. Their definitions are constrained only by the compliance with
SOAP, WSDL, and the XML schema for defining additional “stickers” in the SOAPenvelope.
Comparison Between REST and WS– /ETXPrinciples
Pautasso et al. study. In the most comprehensive comparison to date,
Pautasso et al. (2008 ) compare RESTful and WS– /ETXservices on three levels: (1)
architectural principles, (2) conceptual decisions, and (3) technology decisions.
On the level of architectural principles , Pautasso et al. analyze three principles
(protocol layering, dealing with heterogeneity, and loose coupling) and note thatboth styles support these three principles. However, they can identify only oneaspect common to both styles – loose coupling to location (or dynamic late binding).Consequently, they conclude that it’s not possible to make a decision at this leveland proceed with more detailed analysis. At the level of conceptual decisions ,t h e y
compare nine different decisions and find th at RESTful services require the designer
to make eight of them, vs. only five for WS–
/ETX.H o w e v e r ,W S – /ETXhave many more
alternatives than RESTful services. Finally, in the technology comparison, they
identify ten technologies that are relevant to both styles. In this comparison, WS– /ETX
once again offer many more alternatives than their RESTful counterparts.
Based on these results, the authors recommend using REST for ad hoc integration
and using WS– /ETXfor enterprise-level application integration where transactions,
reliability, and message-level security are critical.
This study illustrates two key difficulties of performing convincing comparisons
of broad ideas, such as Web service styles. First, it’s difficult to select the mostrelevant principles to compare. Second, once the principles are selected, it’s difficultto identify choices that are sha red by the competing ideas.
Pautasso et al. do not explain why they selected protocol layering, dealing
with heterogeneity, and loose coupling as the only architectural principles tocompare. One would expect a comparison of principles to involve non-functionalrequirements ( Bass et al. 2002 ) relevant to Web services. However, in their analysis,
key -ilities (security, reliability) are only mentioned at lowest level of comparison,the technology decisions. Moreover, they shy away from comparing concepts thatare relevant at the enterprise level (transactions, reliability, message-level security),even though they cite these very concepts in their concluding recommendation.

2 REST and Web Services: In Theory and in Practice 39
The actual comparison has two problems. First, they use the numbers of
architectural decisions and available alternatives to choose which style is better.But counting is hardly the right metric – not every decision point has the sameweight. Second, most decisi on points on every level have two options, one for each
style, indicating that they actually have nothing in common. Only in a few cases doboth styles require a decision on the same question. Nevertheless, this paper is thebest-conducted comparison of principles available today. It’s unbiased, thoroughlyresearched, and it examines multiple points of view.
Richardson and Ruby book. A second comparison of note is presented in
the book, “RESTful Web Services” ( Richardson and Ruby 2007 ). The authors,
Richardson and Ruby, discuss the principles that are relevant to all systems availableon the Web. Even though their book is biased toward RESTful Web services, theprinciples they discuss would be a better starting point for making a fair comparisonbetween the two styles.
They identify four system properties of R ESTful services: (1) uniform interface,
(2) addressability, (3) statelessness, and (4) connectedness. In RESTful Webservices, these properties are embodied in resources, URIs, representations, and thelinks between them. Lets consider how these principles apply to WS–
/ETXservices.
Addressability and some f orm of connectedness are embedded in the WSDL
definition of bindings and ports. Many WS-*services are stateless (although it isnot an explicit requirement). Having a uniform interface shared by all services isthe only property not supported by WS–
/ETX. Thus, WS– /ETXservices exhibits three of
these four properties. WS– /ETXservices achieve these prope rties via different means,
but these properties are clearly relevant to both, and therefore a good choice forcomparison.
Richardson and Ruby use a similar approach to evaluate how RESTful Web
services offer capabilities which are important for enterprise-level integration. Theyshow how to implement transactions, reliability, message-level security (conceptsthat Pautasso et al mention, but do not discuss) using REST. We will return to thesethree concepts in “Read y for the Enterprise?”.
Both styles of Web services possess certain characteristics that guide their design
and development, although they are defined in ways that make it difficult to comparethem side-by-side. Next, we will look at how services are used in practice, whichprovides yet another perspective for comparing them.
Survey of Existing Web Services
One obstacle to studying existing Web services is the fact that many of themare not accessible to the outside world, becau se they are proprietary. Proprietary
systems have different requirements (fewer security threats due to well knownvulnerabilities, no need to adhere to common standards) that result in differentchoices of Web services technologies. Industry studies provide some insight about

40 P. Adamczyk et al.
the trends in proprietary Web services, such as the planned and actual usage of
Web services. One industry survey shows that the adoption of SOAP standard byenterprises increased 31% between 2002 and 2003 ( Correia and Cantara 2003 ).
A follow-up survey from 2006 notes that about 12% of enterprises report completinga “full enterprise roll-out” and another 21% are in process, while 60% are stillstudying the feasibility of such projects ( McKendrick 2011 ). Both surveys report
only on WS–
/ETXWeb services.
More recent results show a new trend. According to a 2008 Gartner
Survey ( Sholler 2008 ) there has been an increase in the number of organizations
implementing Web services using Represe ntational State Transfer (REST) and
Plain Old XML (POX). RESTful Web services are considered less complex, requirefewer skills, and have a lower entry cost than WS–
/ETXWeb services. However, the
surveyors believe that RESTful services by themselves do not provide a completeenterprise solution.
Turning our attention to public Web services, two earliest surveys of public Web
services ( Kim and Rosu 2004 ;Fan and Kambhampati 2005 ), from 2004, discussed
strictly WS–
/ETXservices. Both surveys showed that some of WS– /ETXstandards (most
notably SOAP and WSDL) were successfully used in practice, but they did notcover other standards. These surveys have been limited to WS–
/ETXservices, perhaps
unintentionally, because they considered the presence of a WSDL file as a necessary
prerequisite of a valid Web service.
In order to build on their work, we have studied various Web services repositories
(including the only extant ones cited by these surveys) to analyze the availablepublic Web services from the perspective of architectural styles they follow. Weperformed these surveys in mid-2007 and again in mid-2010 by examining the Webservices listed in the following repositories:
xmethods.net
webservicex.net
webservicelist.com
programmableweb.com
These repositories describe only publicly accessible Web services. While SOAP
services are easy to find automatically (b y checking for the presence of the WSDL
file), RESTful services are documented in non-standard ways that make theirautomatic discovery impossible. We exam ined the type of each service manually,
by reading its documentation. We have identified four mutually exclusive categoriesof Web service styles: RESTful, WS–
/ETX,X M L – R P C ,a n dO t h e r .X M L – R P Cw a st h e
first attempt at encoding RPC calls in XML (which later evolved into SOAP). TheOther category groups many other types of services, including RSS feeds, Atom,XMPP, GData, mail transfer protocols. The most popular styles of Web services ineach repository are shown in Table 2.1.
At a first glance, these results could not possibly paint a more inconsistent
picture. Each repository shows a different t rend. However, the differences arise from
the nature/focus of these repositories. The first two repositories, which list (almost)
exclusively WS–
/ETXservices, advertise services that require payment for access. The

2 REST and Web Services: In Theory and in Practice 41
Table 2.1 Web service styles used in public services
xmethods webservicex webservicelist programmableweb
Style 2007 2010 2007 2010 2007 2010 2007 2010
RESTful 3 0 0 0 103 144 180 1627
WS–/ETX 514 382 71 70 233 259 101 368
XML-RPC 10 0 0 6 2 1 2 45 3
Other 0 0 009 8 3 5 9 0 2 0 7
Total
(unique)518
(514)382 71 70 430
(411)459
(386)395
(340)2255
(2179)
Survey conducted in 2007 and 2010. Some service are available in two or more styles. The numberof unique services is shown in parentheses
second repository appears to be closed to registration (we could not find any way
to contact the owners to register a new service) which may imply that they areadvertising only the services which they own. The numbers of services listed inthese two repositories have not changed much in the last 3 years.
The latter two repositories feature a variety of Web service styles, with RESTful
and WS–
/ETXservices being the two most popular styles in both the 2007 and 2010
tally. Programmableweb.com is the only repository that shows an increase in thenumber of services; a fivefold increase over the observed period. Its data showsincrease in all types of services, but mostly in RESTful ones, which currentlyaccount for about 75% of services listed, compared to less than 50% 3 years earlier.
These results, although insufficient to determine conclusively which style is more
popular (and why), indicate that a wide variety of public Web services is availableand that a sizable number of RESTful services has been created recently, even if notall of them are widely known.
REST Concepts in Practice
With so many public Web services available to study, we were able to identify manytrends in how closely services follow the theoretical principles of REST. WS–
/ETX
principles are encoded in XML-based standards that are easy to enforce by tools.The designer selects the necessary features (standards), then finds the tool thatsupports them. The actual development is easy. But since this book is about REST,we will focus on RESTful Web services, and refer to WS–
/ETXonly to compare and
contrast specific features. In this section, we will review how REST principles areembodied and implemented in actual RESTful Web services.
According to the principles of REST, w hich we introduced in “Web Services in
Theory”, every resource is identified with a URI. In response to HTTP messages,resources return their representations to clients, or the clients modify the resources.

42 P. Adamczyk et al.
Fig. 2.1 Examples of RESTful hypermedia defined as URIs. Examples 1–5 use query strings of
formkey=value . Examples 4a–c show alternative ways to define the same resource. Example 6
uses the structural form instead of query strings: the order of keywords is defined by the server’sAPI so that the client need not list keys, only values, in the URI
Proponents of RESTful Web services typically say that every service needs to follow
the CRUD model ( Kilov 1990 ). This concept, borrowed from the database domain,
defines one method for creating, reading, updating, and deleting a resource on theserver (corresponding to POST, GET, PUT, and DELETE methods). This approachenables invoking different operations on a resource by applying a different HTTPmethod. This is only possible if resources are defined in a correct way. Figure 2.1
shows some examples of valid URIs. All of these URIs can be accessed with theGET method.
One good example of a Web service that follows the principles of REST is Ama-
zon S3 (Simple Storage Service) ( Amazon 2011 ). S3 defines many resources and
uses HTTP methods (POST, GET, PUT, DELETE, even HEAD) for manipulatingthem. It uses HTTP error codes correctly and shows how to map various errorsto HTTP codes (the API references 13 unique HTTP status codes in the 300–500range). S3 also supports caching by includi ng ETag header that clients can use in
conditional GET.
However, most RESTful services are not designed as diligently. They neglect
to follow the principles in various ways. In order to evaluate the current level ofunderstanding of REST, we will look at some representative mistakes from theperspective of the 4 principles of REST.

2 REST and Web Services: In Theory and in Practice 43
Identification of Resources
Every designer of a RESTful service must answer the question: What constitutes
the resources of the system? Ideally, any concept within the system that has arepresentation should be exposed as a resource.
In WS–
/ETXservices, clients invoke API methods on the server by passing SOAP
messages to a well-known service end-point defined with a URI. These serviceend-points are the only resources used by WS–
/ETX. Some RESTful service follow
the same pattern – they define one path component to be used in every URI andencode parameters for the corresponding se rver method in the query strings. This is
wrong, because in REST resources are supposed to be accessed with self-descriptivemessages (e.g. HTTP methods) that have well-defined semantics. Looking atExample 5 in Fig. 2.1, it’s OK to access this resource via GET, but what would be
the intended semantics for PUT and other HTTP methods? Such a resource can onlyaccept read-only requests, the way Google’s search service works. But if the clientsneed to be able to modify the resources, this style of resources is not appropriate.
Defining resources is hard. Consider, for example, a hypothetical Web service
that provides information about books and music. Such a service should definemultiple resources, book ,cd,review that are queried by title, author, or ISBN.
Example 1 in Fig. 2.1,example.org/book?title=zen represents a resource
for books that contain “zen” in the title . Examples 2 and 3 show how to query
the resource by author and ISBN. There are several options for defining resourcescorresponding to a review. The system could have one review resource (as in 4a), adedicated resource for each product type (i n 4b), or a composite resource (review)
with individual children resources, one per product type (4c). These all are validchoices. Alternatively, as in Example 6 , the URI structure can enforce a specific
order of parameters (type, then title, t hen author), thus making it unnecessary to
specify the type of each sub-element in the URI. Note that this format requires
implicit understanding of the structure of this URI, which is defined outside of theURI by the provider of this Web service.
The problem of designing resources is sim ilar to teaching object-oriented design
to programmers, who were first taught pro cedural languages – it requires a changed
mindset. One can define resources without deep understanding of REST, but it’sunlikely that such design will take full advantage of all available features of HTTPand URI standards as objects/classes. In the second step, the public methods of theobject are defined. In any non-trivial problem, these two steps identify many objectsand many methods. The application is built by connecting the objects, which invokemethods on one another. A similar approach can be applied to defining resources,except that only the first step identifies ma ny objects (i.e. resources). The available
HTTP methods are defined in the standard and links between resources are traversedat run-time. Thus steps 2 and 3 come for free in HTTP, but only if step 1 is done well.

44 P. Adamczyk et al.
Representations
If resources support multiple representations, they can produce responses in differ-
ent data formats. In HTTP, clients specify their preferred formats in Accept-*
headers for content negotiation. By conforming to HTTP, RESTful Web services
can support multiple types of response (MIME) formats, just like the Web does,which makes it easy to comply with this principle.
Many RESTful Web services support at least two response formats (typically
XML and JSON). Library of Congress Subject Headings Web service is the onlyservice listed at programmableweb.com that advertises the support of contentnegotiation. It serves content in four different types (XHTML with embedded RDFa,JSON, RDF/XML, and N3). Unfortunately, other services do not appear to supportthis important feature of HTTP, because we did not find it in their documentation.
Self-descriptive Messages
REST constrains messages exchanged by components to have self-descriptive (i.e.
standard) definitions in order to support processing of interactions by intermediaries(proxies, gateways). Even though HTTP/1.1 defines eight methods, only two ofthem, GET and POST, have been used extensively on the Web, in part because thesewere the only methods supported by the early Web browsers.
Early RESTful Web services show difficu lties in understanding the differences
between even these two methods.
1Some services defined GET for sending all
requests to resources, even if the requests had side effects. For example, initially,Bloglines, Flickr, and Delicious Web services defined GET for making updatesto these services ( Dare Obsanjo Blog 2011 ). Other services specified that clients
can use GET and POST interchangeably, which is equally wrong. Consequently,these services were misusing Web pr oxies and caches polluting them with non-
cacheable content, because th ese Web systems rely on standard meanings of HTTP
methods. Since then, the offending APIs were modified, but the underlying problemof understanding the semantics of HTTP methods still remains.
Many RESTful proponents consider the use of 4 HTTP methods corresponding
to CRUD operations as a sign of good RESTful design. But these methods are notsufficient to express complex operations on resources. They provide only simpledata-access operations. These methods need to be combined into sequences in orderto implement even the simplest transactions.
2That’s why many RESTful services try
1GET sends data from the server to the client, in the response. POST sends the data from the client
to the server, in the request. Thus, GET is used for reading, and POST for writing.
2A simple bank transaction, e.g. transferring $100 from savings to checking, involves sending
four HTTP requests. First, create a resource for the transfer using POST. Next, send a PUT to the

2 REST and Web Services: In Theory and in Practice 45
to encode more complex operations (such as “search”) into URIs in RPC style even
though they know that it violates REST. Another reason why the CRUD metaphoris not a good match is that HTTP methods POST and PUT do not map exactly toCRUD’s “create” and “update,” respectively. PUT carries a representation producedby the client, which the server should use to replace its contents (so it serves as bothcreate and update). POST means the server decides how to use the representationsubmitted by the client in order to update its resource.
This problem of not taking full advantage of HTTP methods is not unique to Web
services. Typical Web applications (acce ssible via browsers) use only two HTTP
methods in practice. In a study of HTTP compliance of Web servers ( Adamczyk
et al. 2007 ), we found that Web servers and intermediaries understand correctly
only GET and POST methods. Only a fraction of popular websites send compliantresponses to other HTTP methods, even though the popular Web servers implementall these methods correctly. These compliance results haven’t changed much sinceHTTP/1.1 standard has been released, in 1999.
The inclusion of the 4 HTTP methods corresponding to CRUD operations in a
definition of a RESTful service is only a first step in satisfying the principle of self-describing messages. This principle m eans that methods should be used according
to their standard definitions. A case in point is the new HTTP method, PATCH,added in March 2010 ( Dusseault and Snell 2010 ). It is intended to complement
PUT and replace some uses of POST with more precise semantics. With POST, theclient cannot specify how the resource is to be updated. Unfortunately, the definition
of PATCH does not define the structure for including the instructions to update(i.e., patch) the resource. A standard definition of the instructions will be necessaryto make this method interoperable. As the additions of PATCH indicates, the setofrelevant HTTP methods is not static. The WebDA V protocol (which RESTful
proponents tend to overlook) defines 8 more methods for distributed authoring andmanipulating collections of resources ( Goland et al. 1999 ). Thus RESTful Web
services have many self-describing methods to choose from. Although today mostWeb services don’t use their HTTP met hods right, we hope that in time they will.
HATEOAS
Hypermedia as the engine of application state means that neither client nor server
needs to keep the state of the exchange in a session, because all the necessaryinformation is stored in the exchanged HTTP messages (in the URI and theaccompanying HTTP headers and body). Defini ng self-contained links is critical for
RESTful Web services, because these links make it possible to traverse, discover,and connect to other services and applications.
resource specifying the withdrawal of $100 from savings. Then, send a second PUT to deposit
$100 to checking. Finally, send a PUT to commit the transaction. Note that the burden of verifyingthat each step was successful is on the client. If a step fails, the client needs to send a DELETE tothe transaction resource to abort the transaction ( Richardson and Ruby 2007 ).

46 P. Adamczyk et al.
However, this is difficult, because complex interacti ons translate to complex
URIs. Large applications have many states that the client needs to be aware of.HATEOAS forces Web services to expose these states as links, which appear toduplicate the internal implementation of the service. To avoid this duplication, someRESTful Web services resort to exposing the underlying API of the service even ifthey know it’s wrong.
Many services require the client to send user-specific information (e.g. user-id) in
every request URI. As a result, the same r equests from two different clients appear
unique to the Web caches, because caches use URIs as keys for the data. Sendinguser-specific information is often unnece ssary (especially when the user sends a
generic query), but it’s used extensively by Web services providers in order to limitthe number of accesses from each client. S ince HTTP caching cannot be used in this
case (except when the same user requests the same resource again), the service musthandle more requests, which defeats the purpose of rate limiting. This seeminglyinnocuous (but often occurring) lapse violates two principles – the identification ofresources and HATEOAS because the URIs representing states cannot be used byother users. It also affects cacheability.
Other Important Concepts
The HTTP standard defines the meaning of different error conditions and severalmechanisms for caching. Compliant REST ful Web services should follow them.
Initially, RESTful services copied their error-handling mechanism from SOAP.
Many Web services would not use HTTP status codes (e.g. “404 Not Found”)to describe the result of a request, but rather always returns “200 OK” with theactual status is hidden in the response body. Other services (e.g. earlier versions ofYahoo Web services) defined their own status codes that were incompatible withthe standard ones. By using service-specific codes, they would not take advantageof existing Web systems that understand these codes thus forcing clients to buildspecialized, non-interoperable software to handle them. Fortunately, most Webservices we surveyed now do use HTTP stat us codes, and only add service-specific
extensions for new statuses. For example, Delicious uses codes 500 and 999 toindicate that user request was throttled (due to exceeding a pre-defined limit ofconnections). HTTP does not have a statu s that corresponds to this condition, so
it makes sense to define a new one.
Our survey gathered little informatio n about caching. Aside from exceptional
Web services like S3 (and even they don’t use the term caching in the documen-tation), RESTful services do not documen t if they support caching. Of course, the
services that employ user-ids c ould not benefit from caching anyway.
As the length of this section indicates, RESTful services still have difficulty in
following the principles of REST. There are few fully compliant service definitions,but it’s easy to find examples of services th at violate any of the principles. On the
bright side, we have observed a lot of improvements in compliance over the lastfew years. RESTful services, by the virtue of being public are more open to general

2 REST and Web Services: In Theory and in Practice 47
scrutiny. Users can discuss the design decisions in the open, criticize them, and see
changes in the next version. To gain a better perspective of the positive changes thatoccurred over the years, the reader is enc ouraged to browse the discussion of these
and other violations documented at RESTWiki (2011 ).
An important question is: Why are many services that attempt to be RESTful not
compliant with the principles of REST? Are these principles too restrictive? Toohard to implement? Unnecessary for Web services (as opposed to Web pages andWeb applications)? It’s still too early to tell.
Frameworks for Building RESTful Web Services
The improvements in the understanding of the principles of REST, as indicatedby the slow but steady elimination of bad design decisions from public RESTfulWeb services can be attributed to software tools and frameworks that have began toappear in the last few years.
Support of REST Principles
Many frameworks and tools for building RESTful Web services are available today.They are written in different progra mming languages and range from simple to
quite sophisticated in their support of HTTP and other Web technologies. As theycontinue to improve, misunderstandings and violations present in today’s Webservices will likely lessen.
We have examined ten popular frameworks that provide automated support for
building software according to the prin ciples of REST. Some frameworks, like
Ruby on Rails and Spring are generic Web frameworks, while others are specificto RESTful services. Table 2.2summarizes key features of these frameworks,
grouped by REST principles. The frameworks are listed alphabetically, sorted by theprogramming language and name. The second column in the table shows how theseframeworks support defining resources (corresponding to REST principles 1 and 4).Almost all the frameworks provide some support for building resources (URIs) andhyperlinks – through URI templates ( Gregorio et al. 2010 ), annotations in the target
programming language, or other types of mappings. The third column shows whichtypes of multimedia are supported and how ( principle 3). Most frameworks enable
generation of multiple representation formats. The fourth column shows whichHTTP methods are supported (principle 4). Most of them support GET, POST, PUT,and DELETE HTTP methods, either directly, or by specifying the desired method inan auxiliary parameter [such as the X-HTTP-Method
Override header, or the
hidden “ method” form field ( Richardson and Ruby 2007 )]. The last column points
out other interesting features provided by the frameworks. Few brave frameworkshave ventured into implementing more a dvanced concepts of caching, automated
testing, or authentication.

48 P. Adamczyk et al.
Table 2.2 How frameworks for building RESTful Web services support the principles of REST
Name (Prog.
language)Resources and
HATEOAS Representation MessagesOther (API,
caching, statuscodes, etc.)
Jersey (Java) Annotations for URI
mappingsMIME types, XML,
JSON and AtomGET POST PUT
DELETESupport for JAX-RS. Testing
framework
RESTEasy (Java) Annotations for URI
translations andvariable mappingAnnotations for output
representations (manytypes supported).Content negotiationGET POST PUT
DELETEOutput caching and
compression Support forJAX-RS
Restlet (Java) URI templates and
variable bindingSupports various output
representationsGET POST PUT
DELETESupport for JAX-RS.
Caching headers set inConditions class.
Security checks addedvia filters. All HTTPstatus codes
Spring (Java) Templated URIs using
Java annotationsContent negotiation with
Accept header or by
URL inspection (readfile extension)GET and POST
directly, PUT
and DELETEwith_methodETag header for caching
Recess (PHP) URI templates and
variable extractionusing annotationsNot supported GET POST PUT
DELETENot supported
Routes (Python) Proper URL syntax;
No IDs in queryparametersNot supported GET POST PUT
DELETENot supported

2 REST and Web Services: In Theory and in Practice 49CherryPy (Python) Simple mapping:
HTML formsto Python variablesHTML forms GET and
POSTAn object tree generated to
map requests to Pythonfunctions
Django (Python) URI templates for
mapping advancedURL patterns toPython codeTargeted output
formats: XML, JSON,YAMLGET POST
PUTDELETECaching. HTTP status codes
supported by Pythonlibraries
RESTfulie (Ruby,
Java)emphasizes
hypermedialinksmany formats; content
negotiationGET POST
PUTDELETEHTTP status codes;
integrates with Ruby onRails
Ruby on Rails
(Ruby)Route configs map URI
toComponent
class (imposes URIconventions)Excellent support of
many data formats –e.g. Accept headerGET POST
PUTDELETEConditional GET for caching
RESTful authentication

50 P. Adamczyk et al.
Several of the Java frameworks support JAX-RS, a Java API for RESTful Web
services. They are Jersey (considered the r eference implementa tion), Restlet, and
RESTEasy. JAX-RS specifies how to map Java classes to Web resources using Javaannotations. The annotations specify the relative path of the resource (part of theURI) for a Java class, which Java methods correspond to HTTP methods, whichmedia types are accepted by the class, and how to map class properties to selectedHTTP headers ( Hadley and Sandoz 2009 ).
Aside from Django, all the Python and PHP frameworks offer only rudimentary
support for REST. Other frameworks include more advanced features, but they stillfall short of supporting all principles of REST. Most frameworks define schemesfor mapping URIs to classes and methods, but not all of them are as flexible asHTTP requires, e.g. Ruby on Rails imposes constraints on URI formats. Only oneframework (Restlet) supports all HTTP status codes. No framework supports allflavors of HTTP caching, and many do not support caching at all.
The principle of HATEOAS (unambiguous semantics for following and em-
bedding links) is not well supported. Only the RESTfulie framework emphasizesthe importance of this principle. Let’s consider a simple example of the expectedbehavior. When a client requests a resour ce (e.g. information about a collection
of items) it should be easy to construct a URI to refer to an individual item fromthat collection. Frameworks should provide built-in support for such conversions ofURIs. Currently, this mapping work must be implemented manually in the clientcode, because most frameworks do not support it.
Overall, the RESTful frameworks need to include more functionality to be fully
compliant with REST. But the biggest problem is that even if they do implementthe support for a principle, the frameworks have no mechanisms to enforce that it isapplied correctly in the client code.
Ready for the Enterprise?
Frameworks make it possible to build bigger Web services, and their capabilitieskeep on growing. Is that enough to persuade enterprise system architects to switchto RESTful Web services? Recall the study of Web services by Pautasso et al.
(2008 ) we discussed in “Web Services in Theo ry”. They cite security, reliable
messaging and transactions as key differentiators between RESTful and WS–
/ETX
services. To be ready for enterprise, RESTful frameworks need to support thesefeatures. Richardson and Ruby (2007 ) show how these concepts can be implemented
using HTTP.
For basic message-level security, it’s enough to use HTTPS. But more complex
capabilities such as signatures, encryption, or federation (enabling a third party tobroker trust of identities) cannot be supplied by HTTP alone. Further research isrequired to define these concepts properly in RESTful Web services (more aboutthis in “Open Research Problems of RESTful Services”.)

2 REST and Web Services: In Theory and in Practice 51
To provide reliable messaging, one needs to ensure that all HTTP methods are
idempotent. This property makes it possi ble to replay any method, as necessary,
to make sure that it succeeded. Of course, this approach to reliable messaging istedious and currently requires a lot of manual coding on the client side.
Implementing transactions with HTTP messages requires exchanging many
messages, which can get complex quickly (as we saw in “Self-descriptive Mes-sages”). Current frameworks are not mature enough to abstract out/encapsulatecommon transaction patterns. But trans actions are needed as building blocks of
workflows, which occur often in enterprise systems. A proposed extension to theJersey framework introduces action resources for specifying workflows ( Hadley
et al. 2010 ). Each action resource exposes one workflow operation available on
the service. The client obtains the workflow specification (i.e. the list of actionresources) at the beginning of the sequence. In line with the principle of HATEOAS,it’s the client’s responsibility to keep track of the current state of the systemthroughout the execution in order to invoke the workflow resources in the correctorder. This is a dynamic approach, because the exact sequence of the workflowneed not be specified until the client begins to execute it.
But even if security, reliable messaging, and transactions are solved successfully,
RESTful services must also demonstrate scalability. Compared to large legacysystems on top of which many WS–
/ETXservices are built, current RESTful services
are small. Tool support is needed for combining disparate services to build largerones and for automating the generatio n of URI schemas that can adapt when a
service is being extended.
Today’s frameworks are not yet ready to support enterprise needs. They do
not implement advanced security features or transactions; they do not verify that
HTTP methods they generate are idempoten t, which is the necessary prerequisite
for reliable messaging; they are not scalable. Implementing these features is amatter of time, because HTTP already defi nes most of the necessary concepts to
perform these tasks. However, it’s not enough that the frameworks implement thenecessary functionality. The frameworks mus t guide and force the users to recognize
the correct features for the job and to apply them correctly.
Open Research Problems of RESTful Services
REST originated at the int ersection of academia an d software development,
among the architects of the World Wide W eb. Fielding’s research culminated
in authoritative versions of HTTP and URI standards that define the uniquecharacteristics of the Web. Unfortunately , researchers have only recently started
to work on RESTful services. As late as 2007, there were no papers about RESTfulWeb services in either ICWS, ECOWS, or WWW conferences. In 2010, ICWShas featured several papers about RESTful services and the WWW conference hashosted the first “Workshop on RESTful Design (WS-REST 2010)” ( Pautasso et al.
2010 ), which is a welcome sign.

52 P. Adamczyk et al.
Proponents of RESTful Web services made their first attempts to reach the
research community via conference presentations ( Prescod 2002 ;Haas 2005 ),
and computer magazine editorials ( Vinoski 2008 ). Recently, survey papers
(Pautasso et al. 2008 ), and new research work ( Pautasso et al. 2010 ;Overdick 2007 )
began to appear. Hopefully, this book will advance the state of research even farther.
The problems we discuss below are con cerned with non-functional requirements
and how they can be supported by RESTful services. Many of these research effortsare defining new Web standards. Web linking ( Nottingham 2010 ) aims to improve
cache invalidation. HTTP PATCH ( Dusseault and Snell 2010 ) defines a new method
to make more maintainable services. URI templates ( Gregorio et al. 2010 )m a k e
it easier to define groups of resources with regular expressions. OAuth ( Hammer-
Lahav 2010 ) secures authentication and data sharing in HTTP-based systems.
Caching
Of many aspects of performance, cachin g is one of the best examples of why
it pays to use HTTP correctly. The data may be cached by the client, by theserver, or by intermediaries, such as Web proxies. In the early days of mostlystatic content, 24–45% of typical Web traffic was cacheable ( Duska et al. 1997 ).
Today, the estimated range is 20–30% ( Nottingham 2009 ), which is very impressive
considering how dynamic the Web content is.
Unfortunately, most of the RESTful ser vices aren’t benefiting from caching:
many frameworks don’t support caching, an d typical URIs are not cache-friendly,
because RESTful Web services require user info in each request. We have alreadydiscussed how user-ids are used for rate-limiting, in “HATEOAS”. It is unlikely thatWeb services will ever change this policy. Instead, it would be better to move user-specific information out of the URIs, so that the responses can still be cacheable.
An upcoming addition of Web Linking ( Nottingham 2010 ) (for improving cache
invalidation) indicates that the HTTP co mmunity values caching. However, it’s
very difficult to keep up with all the varia tions: caching headers, tags, expirations,
and conditional methods. Caching is so complex that even the upcoming HTTPbisspecification from IETF divides this topic into two documents (Caching proper andConditional Requests). Caches are not unique to the Web: caching in computerarchitecture is understood well. We are lack ing a single, consistent model of caching
on the Web.
Maintainability
Typical maintenance tasks of Web services (adding new features, fixing serviceAPIs) affect services themselves, their documentation, the client code, and eventhe development tools. Since RESTful Web services are still prone to wholesalechanges, each of these facets offers ample opportunities for research.

2 REST and Web Services: In Theory and in Practice 53
Changes of Web service definitions necessitate upgrades of the client code. When
a new version of a service becomes available, clients need to adapt their code.Neither WS–
/ETXnor RESTful services providers are concerned with making client
updates easier. They claim that there is no need to deprecate APIs, because theywill always be available, so clients are not required to upgrade. Ideally, this wouldbe the case, because well-named resources do not need to change ( Berners-Lee
2011 ). New Web services might be able to preserve their APIs for some time, but
maintaining several versions isn’t realistic if a service plans to grow. Some servicesoffer software development tools for building client applications, but they sufferfrom the same types of challenges as typical software – APIs change. Is it time tostart exploring refactoring of Web service APIs?
Security and Privacy
Securing RESTful Web services is a multi-faceted endeavor: it involves securing thedata, as well as the entire communication. One must protect the confidentiality andintegrity of data. The data in transit should be filtered for malicious payload. Thecommunication should support a uthentication and access control, and ensure that
the privacy of the communicating parties is not compromised.
Compared to the WS-Security framework ( Web Service Security WSS ), REST-
ful services rely on various add-ons that work on top of HTTP. HTTPS ( Rescorla
2000 ) is widely used for confidentiality, but it only provides hop-by-hop security.
Developers should adopt message level security mechanisms. Unlike WS–
/ETX,t h e r e
are no standards to follow, but practitioners follow various reference architectures,e.g. Amazon S3 service ( Amazon 2011 ). Amazon S3 also incorporates timestamps
to guard against request replaying. Various client side and server side filters shouldbe employed to validate the content.
HTTP supports basic and digest-based authentication mechanisms ( Franks et al.
1999 ), but both have their weaknesses ( Apache HTTP Server v2.2 2011 ). Current
services delegate identity management and authentication mechanism to a thirdparty, and rely on a claims-based authentication model. Technologies for supportingauthentication for HTTP-based serv ices are emerging, e.g. OpenId ( Fitzpatrick
2005 ) for federated identity, and OAuth 1.0 ( Hammer-Lahav 2010 ) for authenti-
cation and data sharing. These protocols open up new avenues of research. Forexample, OAuth is going through a revision in October 2010, where the protocolwriters are considering dropping cryptographic operations and relying on SSL toprotect plaintext exchange of authentication tokens. They are trading off end-to-endsecurity for ease of programming, and this decision should be validated by research.Another emerging protocol is XAuth ( Meebo Dev Blog 2010 ), an open platform for
extending authenticated user services across the Web, but it still has a lot of opensecurity problems.

54 P. Adamczyk et al.
Fig. 2.2 Security measures adopted at different layers in RESTful systems
Storing URIs in web logs may lead to privacy problems if the logs are not
protected and anonymized. WS– /ETXservices do not store sensitive data in HTTP
method signature and query strings. On the other hand, URIs created for RESTfulWeb services become the audit trail, and they should be anonymized.
Figure 2.2illustrates a hypothetical model of how the security and privacy
measures can be applied together. It shows a secure token service, a key entity ina third party authentication model. Note that the figure does not define the actualsteps of an ideal protocol; it is an open research problem. Researchers also need tofigure out how the security measures fit the REST model.
QoS
When multiple providers offer the same service, a client has a choice and can selectthe most suitable one. Often, this choice comes down to the Quality of Service(QoS) parameters. RESTful Web services today ignore QoS requirements; theironly concern is providing f unctional interfaces. To add QoS parameters to RESTful
services, a language for describing the parameters and a mechanism to incorporatethe description in the HTTP payload is needed. Defining a standard QoS descriptionlanguage might benefit from the work in Semantic Web. Semantic Web ontologiesdefine standard ways of interpreting information, such as QoS parameters, enablingall clients to interpret them the same way.

2 REST and Web Services: In Theory and in Practice 55
Studies of Existing Systems
Web services are good candidates for studying how software engineering concepts
are followed in large, publicly available systems. But there have been few successfulstudies of RESTful services, or side-by-side comparisons of a service that exposestwo interfaces defined in the compe ting styles (one RESTful, one WS–
/ETX).
It is not easy to compare these two styles at the level of principles. The first
order of research is to identify good principles for making the comparison. Zarras
(2004 ) identifies the following principles for comparing middleware infrastructures:
openness, scalability, performance, and distribution transparency. Properties ofsoftware architectures ( Bass et al. 2002 ) is another source of principles to consider.
Another possibility is to apply the same principled approach Fielding used to deriveREST in order to define both RESTful and WS–
/ETXarchitectural styles. This would
entail selecting and applying additional constraints, one at a time, to derive completedefinitions of both architectural styles.
Conclusion
RESTful Web services (and Web services in general) pose the first serious test ofthe principles of REST, as identified by Fielding. On the one hand, the emergenceof RESTful Web services, in response to WS–
/ETXservices can serve as an indication
that REST isthe correct architecture for the Web. On the other hand, the state of
practice still shows gaps in understanding and applying the theory behind REST,
thus indicating that the process is not complete.
Up until a few years ago, there was a simple dichotomy between REST
and WS– /ETX. RESTful services were used only for simple, public services. In
contrast, enterprise standards, tools vendors, and the research community were onlyconcerned with WS–
/ETXservices. This is no longer the case – both styles are being
used in all domains. The new challenge is to use them correctly, and to be ableto align them to solve the real problems o f the enterprise. Can RESTful services
scale up to the enterprise-size challenges? We believe so. Amazon, Google, Yahoo,Microsoft, and other big companies have b een building large, scalable, extensible,
and relatively secure systems on the We b. RESTful services have the same basic
principles to follow.
This concludes our whirlwind overview of how Web services relate to REST, in
theory and in practice. Other chapters in this book will explore these topics in moredetails.
Acknowledgements The authors would like to thank Nicholas Chen, Hakan Erdogmus, Jim
Webber, and anonymous reviewers for reviewing earlier drafts of this work.

56 P. Adamczyk et al.
References
L. Bass and P. Clementes and R. Kazman. Software Architecture in Practice, 2nd Edition . Addison
Wesley, 2002.
P. Adamczyk, M. Hafiz, and R. Johnson. Non-compliant and Proud: A Case Study of HTTP
Compliance, DCS-R-2935. Technical report, University of Illinois, 2007.
Amazon. Amazon Simple Storage Service API Reference, May 2011. http://docs.
amazonwebservices.com/AmazonS3/latest/API/
Apache HTTP Server v2.2. Authentication, authorization and access control, May 2011. http://
httpd.apache.org/docs/2.2/howto/auth.html.
T. Berners-Lee. Cool URIs don’t change, May 2011. http://www.w3.org/Provider/Style/URI.html.
T. Berners-Lee, R. Fielding, and H. Frystyk. RFC 1945: Hypertext Transfer Protocol — HTTP/1.0,
May 1996.
J. Correia and M. Cantara. Gartner sheds light on developer opps in web services. Integration
Developers News , June 2003.
Dare Obsanjo Blog. Misunderstanding REST: A look at the Bloglines, del.icio.us and Flickr APIs,
May 2011. http://www.25hoursaday.com/weblog/PermaLink.aspx?guid=7a2f3df2-83f7-471b-
bbe6-2d8462060263.
B. M. Duska, D. Marwood, and M. J. Freeley. The measured access characteristics of World-Wide-
Web client proxy caches. In USENIX Symposium on Internet Technologies and Systems, USITS ,
1997.
L. Dusseault and J. Snell. RFC 5789: PATCH Method for HTTP, Mar. 2010.J. Fan and S. Kambhampati. A Snapshot of Public Web Services. In SIGMOD Record, Vol. 34, No.
1, Mar. 2005.
R. Fielding. Architectural Styles and the Design of Network-based Software Architectures.
Doctoral dissertation. Technical report, University of California, Irvine, 2000.
R. Fielding, J. Gettys, J. Mogul, H. Frystyk, L. Masinter, P. Leach, and T. Berners-Lee. RFC 2616:
Hypertext Transfer Protocol — HTTP/1.1, June 1999.
B. Fitzpatrick. OpenID, 2005. http://openid.net/.
J. Franks, P. Hallam-Baker, J. Hostetler, S. Lawrence, P. Leach, A. Luotonen, and L. Stewart. RFC
2617: HTTP Authentication: Basic and Digest Access Authentication, June 1999.
J. Garrett. Ajax: A New Approach to Web Applications, Feb. 2005. http://adaptivepath.com/ideas/
essays/archives/000385.php.
Y . Goland, E. J. Whitehead, A. Faizi, S. Carter, and D. Jensen. HTTP Extensions for Distributed
Authoring WebDA V. Internet proposed standard RFC 2518, Feb. 1999.
J. Gregorio, R. Fielding, M. Hadley, and M. Nottingham. URI Template (draft), Mar. 2010.H. Haas. Reconciling Web services and R EST services (Keynote Address). In 3rd IEEE European
Conference on Web Services (ECOWS 2005) , Nov. 2005.
M. Hadley, S. Pericas-Geertsen, and P. Sandoz. Exploring Hypermedia Support in Jersey. In WS-
REST 2010 , Apr. 2010.
M. Hadley and P. Sandoz. JAX-RS: Java API for RESTful Web Services (version 1.1), Sept. 2009.E. Hammer-Lahav. RFC 5849: The OAuth 1.0 Protocol, Apr. 2010.I. Hickson. HTML5: A vocabulary and associated APIs for HTML and XHTML, Oct. 2010.Joe McKendrick. Service Oriented Blog, May 2011. http://www.zdnet.com/blog/service-oriented/?
p0542.
H. Kilov. From semantic to object-oriented data modeling. In First International Conference on
Systems Integration , pages 385–393, 1990.
S. M. Kim and M. Rosu. A Survey of Public Web Services. In WWW 2004 , 2004.
L. Richardson and S. Ruby. RESTful Web Services . O’Reilly , Oct. 2007.
M. Nottingham. HTTP Status Report. In QCon , Apr. 2009.
Meebo Dev Blog. Introducing XAuth, Apr. 2010. http://blog.mee bo.com/?p=2391.
M. Nottingham. Web Linking (draft), May 2010.H. Overdick. Towards resource-oriented BP EL. In C. Pautasso and T. Gschwind, editors, WEWST ,
volume 313. CEUR-WS.org, 2007.

2 REST and Web Services: In Theory and in Practice 57
C. Pautasso, E. Wilde, and A. Marinos. First International Workshop on RESTful Design (WS-
REST 2010), Apr. 2010.
C. Pautasso, O. Zimmermann, and F. Leymann. RESTful Web Services vs. “Big” Web Services:
Making the Right Architectural Decision. In WWW ’08: Proceeding of the 17th international
conference on World Wide Web , pages 805–814, New York, NY, USA, 2008. ACM.
P. Prescod. Roots of the REST/SOAP Debate. In Extreme Markup Languages, EML , 2002.
E. Rescorla. RFC 2818: HTTP over TLS, May 2000.RESTWiki, May 2011. http://rest.blueoxen.net/cgi-bin/wiki.pl.
D. Sholler. 2008 SOA User Survey: Adoption Trends and Characteristics, Sept. 2008.S. Vinoski. Serendipitous reuse. IEEE Internet Computing , 12(1):84–87, 2008.
W3C Working Group Note. Web Services Architecture, May 2011. http://www.w3.org/TR/2004/
NOTE-ws-arch-20040211/.
Web Service Security (WSS). Web Services Security: SOAP Message Security 1.1, Feb. 2006.A. Zarras. A comparison framework for middleware infrastructures. Journal of Object Technology ,
3(5):103–123, 2004.

http://www.springer.com/978-1-4419-8302-2