You are on page 1of 14

Computer Networks 35 (2001) 565±578

www.elsevier.com/locate/comnet

Parlay-based service engineering in a converged


Internet±PSTN environment
Menelaos K. Perdikeas, Iakovos S. Venieris
Telecommunications Laboratory, Department of Electrical and Computer Engineering, National Technical University of Athens,
Heroon Polytechniou 9, 153 73 Zographou, Athens, Greece

Abstract
This paper examines current approaches to signalling and service interworking between Internet and public switched
telephone network (PSTN). It notes that until now convergence between the two networks has for the most part taken
place in the transport and signalling layers. Signalling interworking architectures cater for the speci®c class of tele-
phony-like services and although they can accommodate the extension of the intelligent network's (IN) realm of control
in the Internet, they do not provide a generic platform for service interworking. Through the adoption of the Parlay
APIs, a way is foreseen for (a) consolidation of telephone service over both Internet and PSTN through the imposition
of a uniform call control API while allowing the installed IN infrastructure to be used also for Internet telephony
services and (b) for service interworking between telephony-like services and open distributed services in the Internet.
The paper proposes a service architecture that can be used as a platform for Parlay-based service interworking while
o€setting some drawbacks that the Parlay approach incurs. Ó 2001 Elsevier Science B.V. All rights reserved.

Keywords: Intelligent network; Parlay; Mobile code; Signalling interworking; Internet telephony

1. Introduction trac with adequate quality of service in terms of


latency and jitter. In all these cases, one network's
Convergence between the public switched±cir- service uses the physical infrastructure of another.
cuit networks (SCNs) and the packet-based In- So, convergence at these levels can be viewed less
ternet is a much celebrated trend. However a more portentously as an instance of the encapsulating
careful examination will conclude that at present ability of layered protocol architectures. Problems
this convergence has for the most part taken place at this level revolve around the nature of switching
in the physical, data link and signalling layers. The that takes place and, in general, around quanti®-
ubiquity of local loops connecting subscriber's able characteristics of the medium or of the in-
homes to a now digitised core network renders the formation content. Given that lower layers are
ideal avenue to also connect them with the Inter- usually generic, it is easier to achieve interworking
net. Advancements in quality of service mecha- there than when trying to bridge the semantic gap
nisms in the Internet allow enterprises to use a between di€erent service paradigms and applica-
best-e€ort network to deliver their interoce voice tion-level concepts.
Two patterns of service provisioning can be
easily discerned. One that locates service logic at
E-mail address: perdikea@telecom.ntua.gr (M.K. Perdikeas). the network's endpoints and treats the communi-
1389-1286/01/$ - see front matter Ó 2001 Elsevier Science B.V. All rights reserved.
PII: S 1 3 8 9 - 1 2 8 6 ( 0 0 ) 0 0 1 9 4 - 8
566 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

cations subnetwork as a generic ``get the bytes treme as even these services have some internal
across'' facility and a second one that assumes degree of coherence as they share the concept of a
impoverished terminals and locates service intelli- textual or audio-visual content that ¯ows from a
gence inside the network. Using the SCN's source to reach one or many destinations. Other
switching links to convey Internet trac is not Internet service classes including, for instance,
incompatible with the patterns of service provi- object services that are made available through
sioning that apply in the Internet as Internet ser- distributed, programmatic interfaces are even
vices assume nothing of the intermediate nodes more divergent.
save their ability to route packets correctly. Using There are broad classes of problems that can be
the Internet to o€er telephony services on the other de®ned in this context. The ®rst class is that of
hand, is not so straightforward. For instance, service integration. It relates to the ability of the
switch-based or intelligent network (IN)-based same service to be o€ered by two networks using
telephony services rest on the existence of inter- the same service provisioning structure. Service
mediate nodes and their maintaining state infor- integration therefore seeks to integrate two origi-
mation for the duration of a call which is nally distinct service models into a single one thus
something not granted for free in a connectionless facilitating seamless service provisioning across
network. di€erent networks. The second class of problems
Supplementary, switch-based services were the relates to questions of service interworking.
®rst that exploited the fact that call state was The remainder of this paper is structured as
maintained by intermediary switches. The switch follows. Section 2 examines practices for service
thus became not only an instrument for steering integration for the particular class of telephony
the modulated waveforms or, later, the digitised services. Signalling interworking is shown to be
frames but also the point of service provisioning. It narrower in scope and a sort of preamble for the
has to be noted at this point that the SCN concept support of a more thorough interworking at the
of signalling corresponds to a protocol pattern service level which is covered next. We then pro-
which is not found in the Internet. Signalling is ceed to examine service interworking and describe
terminated, examined and possibly operated upon the Parlay approach which is the most compre-
and ®nally reinstated at each intermediate node. hensive cross-network service framework stan-
This allows partial, yet synchronised, images of a dardised so far and we present an architecture that
call to be maintained by an array of network utilises distributed object and mobile code tech-
switches. This is in sharp contrast with the Internet nologies to compensate for certain drawbacks that
where although the network protocol is indeed Parlay incurs.
terminated and reinstated at each router, it is at a
level of abstraction far below that of signalling and
has no service-related concepts built into it. Indeed 2. Consolidation of the telephony service
telephony signalling is modelled as an application
layer protocol in Internet applications. It is the The public switched telephone network (PSTN)
fact that an application-layer protocol is observed is geared towards the provision of the plain old
by intermediate entities instead of ¯owing trans- telephone service. When augmented with an ISDN
parently through them that has enabled the pro- or ATM bearer capability, it can also be used
vision of switch-based and, later, IN-based services to deliver multimedia services like, for instance,
and renders simple interworking schemes unable video-telephony or video-conferencing but even
to support a common service model for both In- these services can be viewed as enhanced versions
ternet and SCN-based telephony services. of telephony service as they share many concepts
Moreover, the Internet has given rise to services with it. For the remainder of this paper when re-
that are incompatible with the model of telephony ferring to telephony services, we will implicitly
services. The Web browsing and e-mail services are include these services as well. It is this class of
just two examples and not among the most ex- services that characterises SCNs and whose
M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578 567

seamless provisioning over packet-based networks speci®ed session initiation protocol (SIP) [11].
is sought. Since IN [3] is used to implement the There are thus three main architectures for Inter-
more control rich services (e.g., video-conference) net telephony: the H.323, the MEGACO and the
and a lot of supplementary features of others (e.g., SIP one, not excluding the possibility for hybrid
three-party calls) the litmus test for uniform pro- architectures. These di€er mainly in the nomen-
vision of telephony services over the Internet and clature they use for the Internet-resident telephony
the PSTN is the ability to use the same IN super- entities they de®ne (generically referred to in this
structure over both the networks. This will impose paper as ``soft switches''), the roles they assign to
a uniform service provisioning model and allow them and their scope.
telephony services to be o€ered in a truly consol- The purpose of this paper is not to provide a
idated manner. detailed discussion of the characteristics of the
Voice over Internet protocol (VoIP) refers to various VoIP architectures. Our aim is mainly to
the transmission of digitised voice as the payload show how these Internet telephony standards
of Internet protocol (IP) packets. Before the ac- ®t into an overall architecture for service inter-
tual transmission channel can be established working. So we will brie¯y describe the architec-
however, signalling between the two involved ture of an H.323 Internet telephony system and
parties must take place. Signalling must also be will underline its commonalities with other VoIP
employed during the call (e.g., for call transfer or architectures particularly with regard to how they
to invite another party) and for gracefully ter- function as parts of a broader service interworking
minating the connection and releasing any framework.
reserved resources. Internet telephony signalling The H.323 is an umbrella speci®cation that de-
has the same objectives as PSTN signalling scribes terminals and other entities that provide
di€ering only in that it is an application layer multimedia communications services over packet-
protocol and not an intrinsic capability of the based networks which may not provide a guaran-
network infrastructure. This gives it great poten- teed quality of service. In general, H.323 provides
tial for evolvement (being more software-based) the foundation for audio, video and data com-
but also detaches it from common patterns of munications across the Internet. The scope of the
service provisioning in PSTN that rested on the standard however, does not include the network
availability of network maintained call state interface or the transport protocol used on the
information. network and so H.323 can be used in other net-
works as well. H.323 is broad in scope as it
2.1. Signalling interworking addresses codec standards, multicast support,
bandwidth management and call control aspects.
Telephony services have the greatest internal Fig. 1 depicts the H.323 architecture and the main
degree of coherence and so before attempting to entities that participate.
pursue any further-reaching interworking models The role of the H.323 gateway is to re¯ect the
between Internet and SCN services, it is rational properties of one network's endpoint to another.
that we should ®rst seek to de®ne a comprehensive So, the H.323 gateway terminates and reinstates all
framework for cross-network provision of the signalling layers at both sides (and mediates be-
telephony service. Once the telephony service is tween them) and is responsible for translating be-
consolidated under a single service provisioning tween media transmission formats as well.
model, broader schemes can be advanced. Prob- Naturally, the gateway is optional in cases where
lems of interworking in a more general context terminals of the same type are used. The gate-
may be deferred until the simpler to tackle cases keeper is another optional entity but nevertheless
have been dealt with. an important one when it is present. It provides
Internet has de®ned its own standards for tele- address translation, call control services and
phony signalling including ITU-T's H.323 suite bandwidth management for terminals and gate-
[12], MEGACO [9] and the lightweight IETF- ways in its zone.
568 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

Fig. 1. H.323 architecture and the main entities participating.

The MEGACO model is similar to that of H.323 ently over di€erent lower layers (using layer ad-
although more modular and decomposed. SIP on aptation at a low level). On the other hand, when
the other hand is a lightweight text-based protocol two di€erent signalling architectures exist, signal-
for session initiation. Like H.323 it contains ling mediation is necessary which means that the
primitives for call and connection control but of gateway terminates all signalling layers and un-
reduced articulateness. dertakes a process of syntactic and semantic
There are two conceptually di€erent approaches translation between the forms and procedures used
to interwork between terminals of di€erent types in the one signalling protocol and those used in the
and/or located in di€erent networks which both other (as the H.323 gateway does).
confusingly bear the name ``signalling interworking''. The ®rst approach is speci®ed by the SIGTRAN
When the same signalling standard is imposed but working group of IETF and uses the Internet as a
di€erent network technologies are used, the upper trunk to convey the upper part of the signalling
signalling layers are not terminated at the location system 7 (SS7) stack on top of it. It is depicted in
of a gateway but are instead conveyed transpar- Fig. 2. The main concern here is SS7 connectivity

Fig. 2. Complete signalling encapsulation. Only the lowermost singling layers are terminated ± the others are conveyed transparently
as IP payloads. The same signalling protocol extends uninterrupted over both networks.
M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578 569

and so this approach quali®es for a low layer with this architecture, while every public switch in
interworking scheme. Consequently, it is mostly the PSTN would be able to support a service
relevant to carriers who want to use a private IP switching function (SSF), similar capability would
network to convey their SS7 trac. As an exam- not be found in the Internet side as all call state
ple, there are at least two Internet drafts [10,14] information would be maintained in the end-user
proposing to adapt the SS7 stack onto the Inter- terminals and not in network-resident entities. If
net. The ®rst provides an adaptation layer at signalling information is observed only at the ter-
message transfer part 2 while the other at message minal-located media gateway, then any e€ort to
transfer part 3. This approach is also used in a extend IN's realm of control in the Internet is in-
local scope between the MEGACO signalling validated. The other drawback of the previous
gateway and the MEGACO media gateway (which solution is that it ties Internet telephony with
is the actual signalling termination point). The PSTN signalling standards which is not appro-
MEGACO signalling gateway does not mediate priate as it fails to cater for the speci®c needs of
between di€erent signalling protocols and only Internet telephony applications and forfeits their
takes care to convey Q.931 signalling transparently evolutionary prospects due to their software
over IP to the location of the media gateway which nature. Of course, these two drawbacks are not
is the real terminating point (and can be imple- intrinsic to the MEGACO architecture in general ±
mented inside the terminals). Under this approach they relate only to this particular deployment.
(a terminal-implemented media gateway), we have MEGACO is modular and can support con®gu-
PSTN and Internet terminals using the same sig- rations which su€er from neither of these two
nalling standard. problems.
While using the same signalling protocol over What the previous discussion achieved is to
both networks evades the more dicult problem make explicit that the target should be such that
of signalling interworking at a higher layer, the each domain should get to keep its own standards
main problem with this approach is that on the (thus not impeding evolution) and that Internet
Internet side, PSTN signalling is completely en- telephony signalling should be observable at some
capsulated inside IP packets. Accordingly, on the points. This brings us to the second approach, in
Internet side, all functionality and state informa- which PSTN signalling is terminated at the gate-
tion are moved to the terminals. This is of course way and then converted to Internet telephony
consistent with the Internet approach but makes it protocols. This is re¯ected in Fig. 3 and adopted
very dicult to deliver an integrated telephony by the H.323 gateway.
architecture particularly with respect to its ability Under this approach, di€erent signalling stan-
to exploit the IN infrastructure. For instance, if dards are used in the two networks and explicit
one would try to use an existing IN infrastructure interworking between them takes place. To satisfy

Fig. 3. All signalling layers are terminated, the uppermost are converted to Internet telephony protocols.
570 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

the second requirement (of some network-side be reported to active IN service logic instances and
control of the exchanged signalling messages), it is it is also the BCSM that converts INAP instruc-
fortunate that all Internet signalling standards can tions received from the SCF to call signalling
impose the requirement that at least call signalling messages emitted from the switch.
is routed through intermediate entities with the In all the available Internet telephony signal-
sole intent that these entities maintain call state ling standards, a call state model is either de®ned
information and allow a service logic to operate on explicitly or implicitly by the sequencing of mes-
it according to the IN's listener±controller pattern. sages and the response of the signalling endpoints
Such modes of operation are commonly referred to them at di€erent times. So state models similar
to as ``gatekeeper routed signalling''. Once termi- to the BCSM can be identi®ed in the terminals
nals are instructed to have signalling routed participating in an Internet telephony session. As
through network entities, maintenance of call state indicated in the previous section, by routing the
by network entities is accommodated and appro- signalling through network-resident entities with
priate entities that have access to enough runtime the sole purpose of allowing them to hold state
signalling information can be allocated to perform information, these state models can become sim-
the SSF role according to the IN model. The IN ilar in functionality to the BCSM and become
pattern of service provisioning and the role as- IN's point of service provisioning for Internet tele-
sumed by the SSF are examined in Section 3. phony services as well. What is needed is the
de®nition of the appropriate correspondence be-
tween these state models. As long as this corre-
2.2. IN-based service provisioning spondence has been de®ned [8], H.323,
MEGACO and SIP soft switches can be viewed
The IN model of service provisioning utilises a for IN purposes as SSF entities and IN services
very common listener±controller pattern. IN ser- will simply observe underlying SSFs without be-
vices are executed inside a service control function ing able to distinguish which of them are imple-
(SCF). The SCF receives noti®cations from and mented on top of ``real'' PSTN switches and
issues commands to the SSF using the intelligent which on top of Internet soft switches. Note at
network application protocol (INAP) [13]. The this point that H.323 or SIP call models can
semantics of the messages exchanged over the support SSF emulation at either the BCSM or the
SCF±SSF interface reveal that an abstraction of IN-SSM level. Call control signalling is routed
the underlying network resources is o€ered to the through these soft switches to allow state to be
SCF (and hence, to the services). This abstraction maintained and operated upon by the SCF-lo-
takes the form of a basic call state model (BCSM) cated service logic programs. Thus, in this man-
which is implemented in the call control function ner a single IN-based service provisioning
of the network switch. In IN capability set 2, there superstructure can be applied seamlessly to
is an even higher level abstraction o€ered on top of underlying networks and the same service model
the BCSM called the IN switch state model (IN- will be used to provide telephony services over
SSM) but that does not alter the model of service circuit- and packet-based networks. Fig. 4 depicts
provisioning in any way. Commands issued by the this. A variety of interworking arrangements for
services towards the SSF are ultimately translated signalling may be present although not explicitly
into more ®ne-grained instructions used to mani- indicated in order to avoid cluttering. The point
pulate the BCSM and in the opposite direction is that uniformity at the IN control level is not
events originating from the BCSM are presented inconsistent with a diverse set of signalling pro-
to the service logic programs in the form of the tocols employed underneath. Observe that the
messages that are available at the SCF±SSF in- SCF is common for both networks and that
terface [1]. It is the BCSM that detects basic call SIGTRAN is used to transparently convey the
and connection control events that can lead to the INAP ¯ows to the Internet-based soft switches
invocation of IN service logic instances or should that support the SSF functionality.
M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578 571

Fig. 4. Application of the IN service model over the Internet with the use of soft switches-routed signalling.

3. Service interworking semantically narrow interface. The scope of service


provisioning is local (i.e., it applies at the origi-
Service interworking is di€erent in its perception nating or the terminating exchange) and the ap-
from service integration in that it does not view erture of visibility provided to the services by the
services as self-contained bodies of functionality switch is limited and requiring transactions to be
that are provided to end-users but as open pieces established through complex protocol stacks. An
of code, composed of, or having access to more SS7 network is used to convey the INAP protocol
rudimentary facilities they might wish to share ¯ows that are established between the various IN
with other services or reciprocally to leverage on functional entities, the most important of which is
facilities provided by others. The PINT standard the SCF±SSF ¯ow.
[16] for the click-on services is one of the few in- IN is said to be closed in the sense that although
stances of service interworking standardised so far. an event noti®cation and enforcement point is
It allows a cgi-bin program triggered in response made available at the SSF interface, that interface
to a user clicking on a hyperlink on his browser to cannot be exploited by other service providers.
request from the PSTN the setting up of a call. Because of these restrictions, IN's full potential is
PINT assumes a third-party initiated call capa- not reached and it itself comes close to being
bility and can be used in connection with a SIP simply a more ¯exible implementation alternative
architecture but it provides a very limited and compared to switch-based services.
basically ``request-only'' handling of functionality In a world where a variety of terminals is
in the PSTN. In short, PINT amounts to little available, where service subscribers want unim-
more than an ad hoc arrangement and falls short peded personal and terminal mobility and where a
of providing a comprehensive framework for ser- multitude of event sources and possible handles
vice interworking. for action are available, such ``con®ned'' service
models are no longer appropriate. The Internet
3.1. Open service provisioning has brought with it a plethora of services with
vastly divergent characteristics compared with the
The IN model for a service is one of an opaque relative homogeneity that characterises PSTN
body of code that manipulates call processing on a services. In any intended service interworking
set of subordinate switches using a powerful yet scheme, how will appropriate metaphors be sought
572 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

to allow interaction between di€erent service par- An API-based approach has been adopted by
adigms? the Parlay group [15] in its speci®cation for the
Distributed object technologies and the re- various services it deals with. The goal in Parlay is
placement of message-based protocols with ap- to de®ne open, object-oriented interfaces that will
plication programming interfaces (APIs) o€er a expose network functionality to edge-of-network's
way out of many of the problems impeding ser- applications. While IN simply disentangled the
vice interworking. Essentially the goal is to create service switching from the service control function,
a uniform service space or layer where services Parlay moves a step further by allowing them to be
can be executed. Execution of a service entails the undertaken by di€erent actors. By providing edge-
availability of programmatic hooks underneath of-network application developers with the ap-
and of noti®cation points so that a service can be propriate interfaces with which to exert control
kept informed of events of interest and synchro- inside the network, the task of service creation
nise its execution with their occurrences (see [4] becomes no di€erent than that of application de-
for a typical approach). Therefore the IN listen- velopment. Moreover, the same tools and meth-
er±controller pattern where a physical entity (in odologies can be used in both cases as said
IN, a switch) exposes part of its functionality to interfaces are based on distributed processing
be controlled by a remotely located entity which technologies and can be readily mapped on the
it also undertakes to inform of events for which more popular relative standards including the
the latter has registered interest, can be extended common object request broker architecture
to arbitrary services. Given the diculty in ab- (CORBA), the distributed component object
stracting from the diverse set of services that can model (DCOM) and Java's remote method invo-
be found (or can be introduced) in a converged cation (RMI). In this manner, a resource pool that
Internet±PSTN environment, this pattern is was yet untapped for network operators ± that of
powerful and possibly the most generic scheme application programmers ± can be utilised for the
®tting all services. Essentially the application of purposes of network services development thus
this pattern across services coming from di€erent furthering the convergence between the two in-
network technologies introduces a separation dustries in the aspects of the paradigms of soft-
between the controllable and controlling parts. In ware development that are used and the tools,
a second step, the application of distributed ob- processes and methodologies that are involved.
ject technologies enables the spatial separation of Alternatively, a network operator can outsource
the two parts and it further allows this to happen the task of writing services for his network and
in a location-transparent manner. This could not concentrate on his core competencies. Knowledge
happen with SS7 as it does not display the of complex protocol stacks and esoteric features
properties of a distributed processing environ- and procedures used in telecommunications ceases
ment (DPE). to be a prerequisite for writing services and an
A further bene®t accruing from the application understanding of the semantics of an API is all
of distributed object technologies is that through that is required. The relative coarse granularity of
the introduction of a common object model, many the Parlay APIs further contributes to both the
inconsistencies that would have otherwise ham- ease with which to build services using them and
pered such an interworking e€ort are removed. the ability of the network to protect its own in-
Such inconsistencies could have been related to tegrity (which might have been more easily com-
di€erent wire formats (marshalling speci®cations promised if lower level and more elaborate handles
and procedures, data structure de®nitions), direc- were exported). As a matter of fact, establishing
tory mechanisms and life cycle operations. Also, it that access by application vendors to the net-
is much easier to guarantee compliance with an work's functionality will not compromise its in-
API than it is with a protocol (in DPE imple- tegrity is also the responsibility of appropriate
mentations, this is simply an instance of runtime horizontal authentication and authorisation in-
type checking which is done for free). terfaces.
M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578 573

3.2. The Parlay interfaces speci®cations include APIs for, among others, call
control, messaging, user interfacing (in the IN in-
The Parlay speci®cations (now in version 2.1) telligent peripheral sense), connectivity manage-
de®ne APIs designed to enable network operators ment and mobility management. A short
and independent software vendors to provide ser- description of the most important of these APIs
vices across Internet, PSTN and wireless networks. follows.
As such, the scope of the speci®cations is quite The call control speci®cation de®nes a generic
broad. The general framework for the Parlay ser- call control service, an INAP-1 call control service,
vices and the business model it implies can be an enhanced call control service and multimedia
represented by the ternary relationship depicted in and conferencing call control services. The generic
Fig. 5(b). service provides the more rudimentary form of a
As can be seen from that ®gure, the IN does not call control service. It is based around a third-
modify the business model of a simple subscriber± party model, which allows calls to be instantiated
network provider relationship as the service provi- from the network and routed through the network.
sioning function remains within the realm of the The intent is to support enough functionality to
network operator. It is the externalisation of the allow call routing and call management for con-
``C'' interface that succeeds in exporting this func- temporaneous services. On top of this generic call
tionality in an open competitive market of many control service, the INAP-1 call control services
service providers. In doing so, it enables a subscriber provides additional functionality resulting in a call
to independently modify his relationship with a state model similar to that of IN capability set 1.
number of service providers while attached to the More sophisticated applications can thus be de-
same network operator. Clearly, this creates a more veloped although full leg management is not sup-
¯uid market than the mere ability of a customer to ported. The enhanced call control service adds leg
subscribe to the additional service features that his management to the capabilities o€ered by the ge-
carrier introduces from time to time. neric call control service. It also adds some load
There are two more reasons that attest to Par- control features. The multimedia, conferencing
lay's superiority over more closed models of ser- and multimedia conferencing call control services
vice provisioning. The ®rst is that in contrast to extend the enhanced call control service to allow
IN, Parlay's focus is not only on switching systems control over the media channels associated with
but on a broader set of equipment that can be the legs in a multimedia call, creation of confer-
described as ``network resident''. The relative ences, moving of legs between sub-conferences,

Fig. 5. The IN (I) and the Parlay (II) business models.


574 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

Fig. 6. Parlay is broader in scope but uses the same listener±controller model as IN.

merging of subconferences and handling of mul- the Parlay framework by providing it with the
timedia conference policies. pointers to those callback methods. This corre-
The implementation of these call control APIs sponds to the well known listener±controller pat-
will naturally have to rely on concrete call control tern by which an object registers itself as a listener
technologies. However, the APIs themselves are to an event source so that it can receive noti®ca-
cross-network and technology independent mean- tions. The controller can then use the hooks pro-
ing that the scope of service provisioning extends vided by the event source to e€ect some behaviour.
uninterrupted over both circuit-switched and The relationship between a Parlay application and
packet-based networks. Interworking mechanisms the Parlay service components is thus markedly
such as those described in previous sections will similar to the relationship between the service
have to be used but this is a concern of the net- switching and the service control functions in IN
work provider and not of the application devel- only that here methods are used instead of pro-
oper that uses the Parlay interfaces. From the tocols and the services have a much broader scope
point of view of the application, di€erences be- of control. Since the APIs are provided over a
tween underlying network technologies are not distributed infrastructure this also carries with it
perceived thus truly imposing an uni®ed call con- the additional bene®ts of more ¯exible service de-
trol model. Fig. 6(b) depicts in more detail the ployment and of a laxer binding between the ac-
available Parlay APIs on the network side and tual physical systems.
their relationships.
Note that according to Parlay, edge-of-network
applications should also provide some interfaces of 4. A Parlay-based service architecture
their own which are used by the network-located
service components as callbacks for noti®cations, The approach adopted by Parlay re¯ects the
event reports, etc. The applications interact with understanding that services are ever moving out-
M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578 575

wards. From switches to service control points and ing the operator±service provider relationship with
now, to enterprise servers at the edge of the net- any additional semantics. Attaching protocols and
work. There are, however, two issues related to runtime associations (in addition to the already
this locating of functionality at the network's edge. de®ned APIs) to it will back®re on Parlay's main
One has to do with the possible performance goal. It should be intuitive that in order to tackle
degradation as services reside at the network's this problem and still observe the aforementioned
periphery connected with it through a best-e€ort requirement only generic, middleware-based en-
IP network (commercial DPEs run over IP). The hancements ought to be sought. We hold that
other relates with the support of personal mobility mobile code technologies can be employed to that
and the ability of a user to register for a subscribed end [2]. A mobile code infrastructure is generic in
service at any terminal in the network, provided purpose as it o€ers a mere ability to transfer
that the terminal has the required capabilities. software objects. Being generic, the installation of
Personal mobility provides the user with a single appropriate middleware between the service pro-
point of contact for the various modes of com- viders and the network operator will still maintain
munication and allows him to enjoy the same the looseness of their coupling. A service provider
portfolio of services regardless of the network in on the other hand will be granted the ability to
which he is roaming. The separation imposed by inject his services into the operator's core network.
Parlay between the roles of the network and the The business model of this relationship is more
service provider makes functional aspects of per- involved than that, of course, but conceptually it is
sonal mobility easier to support as the roaming of a lease model whereby the service provider pays
a user a€ects the network provider he uses and can for the processing power and communication re-
be tackled independently from the issue of unin- sources his services use while executing in the
terrupted service provisioning over multiple net- network operator's space.
work providers. The localisation of services, The STARLITE project [5,18] aims at an ar-
however, at the edge of the network makes it dif- chitecture that can deliver the full functional po-
®cult to guarantee optimal access to services or to tential of Parlay while also compensating for its
integrate Parlay-type service provisioning with identi®ed drawbacks. The approach adopted in
hierarchical, tree-like, network-wide models of STARLITE is to leverage on mobile code capa-
service provisioning and deployment. The problem bilities as suggested above. Viewing services as
is even more evident when considering the hand- mobile code objects leads to an important di€er-
ling of mobility management data where such a entiation with the Parlay model. The view that
hierarchical distribution and replication is widely Parlay takes of the application is that of an opaque
practised. Enterprise-held data, although easily piece of code. This is why the speci®cation e€ort
accessible by edge-of network provided services, is focuses narrowly on the de®nition of a set of APIs
not easily integrated into such distribution chan- in order to grant enterprise-resident applications
nels. The need therefore arises to reconcile the access to network-centric functionality. Network
Parlay model with patterns of service provisioning services in the STARLITE system are not, how-
which view the services and their data as ubiqui- ever, opaque pieces of code executed at the net-
tously available throughout the network. work's periphery. Since STARLITE services are
One of the main successes of Parlay is that it mobile objects, a need arises for an underlying
allows fully functional service provisioning while architecture that can provide structured access,
making the coupling between service providers and deployment and container-related operations for
network operators very loose. This property must its services; more needs to be known about the
be respected by any architecture aiming at non- services that make the Parlay calls to the network
functional improvements. What is needed is the if management and lifecycle operations on them
ability to keep service creation at the edge of the are to be handled eciently. Note here that since
network and maintain the runtime provisioning of commercially available mobile code platforms are
services from inside the network without burden- Java based, services are also implemented in Java.
576 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

What is necessary is the de®nition of a component engaged in communication with each other when
object model for Parlay services. Once such model in fact the interposed middleware mediates method
is available, any supporting architecture will be invocations between the two. Thus, the functional
able to transparently take care of issues like rep- association between the services and the network is
lication, migration, activation and passivation. not burdened with additional relationships and the
Services will be then executed in specialised con- service provider and network operator can still be
tainer environments where they will enjoy the two independent actors. The STARLITE network
services mentioned above plus access to the Parlay is responsible for brokeraging between the two.
interfaces exposed by an underlying network. The This is depicted in Fig. 7. An added bene®t of the
described system essentially constitutes a three-tier use of the execution contexts or containers as de-
architecture with the STARLITE network being picted in Fig. 7 is that by supporting proxies for
the middle tier. Thus, the typical two-tier Parlay the actual network-side Parlay objects, services can
architecture is augmented with an additional be pure Java objects delegating the responsibility
mobile code-based tier that interposes itself be- of using the appropriate middleware technology
tween the network and the service providers for communication with the Parlay components
bringing with it the operational enhancements (DCOM, CORBA or RMI) to the containers. A
identi®ed above. However, this tier is transparent common object model de®ned in Parlay is used for
for both the network and the services themselves all underlying equipment at the DPE middleware
who maintain the illusion that they are directly level. Similarly, a common Java object model is

Fig. 7. A Parlay-based architecture.


M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578 577

used at the proxy level for all network resources Functionally, the architecture depicted in Fig. 7
that are represented facilitating cross-network preserves all of Parlay's bene®ts while annulling
service creation. some of its more awkward implementation charac-
The middle tier provides the appropriate execu- teristics. For the middleware infrastructure in the
tion environments into which services developed by middle tier to be able to undertake the functions
independent software vendors will execute. These necessary in the life cycle management of the mobile
execution environments can be powerful enterprise code it hosts, and still support the ability of the
servers allowing further economies of scale to de- services (and of the network code on the other side)
velop as hundreds of services will be executed there. to view only Parlay APIs, certain interfaces have to
Service providers will still have the option to keep be de®ned separating the views that the network
service execution within their premises as the Par- provider, the service provider and the infrastructure
lay APIs are implemented over a DPE making the have of the services. In our architecture, we have
actual runtime location of the services insigni®cant. de®ned appropriate interfaces in Java to allow this
For services, however, that impose rigorous stan- separation of views to take e€ect. Services being
dards of performance, use of the operator-provided Java objects implement all the relative interfaces
execution environments can be obtained. From the and so polymorphically expose di€erent facets to
point of view of the network provider on the other di€erent components in their environment. The
hand, no additional risk is incurred due to the co- service distribution manager has access to migra-
location of externally developed services with crit- tion-related methods while the execution contexts
ical network equipment as exactly the same remote are able to invoke serialisation and de-serialisation
interfaces are used in both cases. methods on services in order to move inactive ser-
Once services are injected inside the network, vices from memory into persistent storage and then
runtime mechanisms can decide transparently as to back into memory when an incoming request
their optimal location or to facilitate access to the arrives. Finally, Parlay components on the ®rst-tier
replicated data they may require. Possible move- see only Parlay-de®ned methods corresponding to
ment of services from one execution context to the services' callback interfaces.
another will be unobserved by the network oper-
ator as it will be taken care of by the mobile code
middleware. Mobile code middleware is the ideal 5. Conclusions
means to ensure that the network operator will
remain oblivious to performance-driven optimisa- The real breakthrough in the Parlay approach is
tions and that no additional functional associa- that of allowing the handling of an asynchronous
tions are introduced between the network and the telecommunications event to be carried out in an
service provider. Relative, commercially available execution context outside the con®nes of the net-
middleware (e.g., the Grasshopper platform [17]) work, in an area outside the administrative or
transport both code and transient service state management responsibility of the network opera-
information in a generic manner allowing unim- tor. For the ®rst time this makes it possible for a
peded service logic mobility without the need for thread of control that was spawned inside a tele-
any service-speci®c provisions. Services can roam communications network to enter into processing
between the network operator-provided execution at the public Internet. This is the fundamental
contexts or even between execution contexts pro- achievement. Once this is allowed, all other bene-
vided by di€erent network operators (possible ®ts of Parlay (e.g., those related to cross-network
following mobile users) without the latter per- creation of services) accrue naturally and cost-free
ceiving it. Code migrating operations may be simply by virtue of the fact that code now executes
prompted either by the need to optimise load on the Internet side and can swiftly take advantage
balancing in the use of both processing and com- of APIs that are made available there and which
munication resources [6] or in response to personal are typically less rigidly standardised and usually
mobility events [7]. destined to span a shorter life cycle.
578 M.K. Perdikeas, I.S. Venieris / Computer Networks 35 (2001) 565±578

The locating of services outside the network sits [7] M. Breugst et al., Impacts of mobile agent technology on
oddly however with standardised approaches for mobile communications system evolution, IEEE Personal
Communication Magazine 5 (4) (1998).
the location, replication and in general, handling [8] D. Ackermann, J. Chapron, Is the IN call model still valid
of mobility data or service instances across care- for new network technologies? in: Proceedings of the ICIN
fully de®ned hierarchies over wide geographical 2000, Sixth International Conference on Intelligence in
areas. The prospect of vastly personalised services Networks, Bordaux, France, 2000.
(opened up by the much more diversi®ed envi- [9] F. Cuervo et al., Megaco Protocol version 0.8, IETF RFC
2885.
ronment in which Parlay services operate) em- [10] G. Sidebottom et al., SS7 MTP3-User Adaptation Layer
phasises this shortcoming. Mobile code technology (M3UA), IETF draft.
can be used on a middleware basis to o€er a [11] M. Handley et al., SIP: Session Initiation Protocol, IETF
platform for data or code mobility that is generic draft.
and thus introduces no associations that could [12] ITU-T Rec. H.323, Packet-based Multimedia Communi-
cations Systems.
back®re on the success of the primary goal pur- [13] ITU-T Rec. Q.1224, Distributed Functional Plane for
sued by Parlay, namely that of separating the role Intelligent Network Capability Set 2.
of the service provider from that of the network [14] K. Morneault et al., SS7 MTP2-User Adaptation Layer,
administrator. The IST STARLITE project in IETF draft.
which the authors participate has undertaken the [15] Parlay APIs 2.1, Generic Call Control Service Interfaces,
available from www.parlay.org.
design and implementation of such an architec- [16] S. Petrack et al., The PINT Service Protocol: Extensions to
ture. SIP and SDP for IP Access to Telephone Call Services,
IETF RFC 2848.
[17] Grasshopper Agent Platform, http://www.ikv.de/products/
Acknowledgements grasshopper.
[18] Services Technologies and Architectures Enabling Internet
and PSTN Integration, http://www.ist-starlite.net/.
This work has been partially funded by the EU
IST project STARLITE and the Greek GSRT
Menelaos K. Perdikeas received the Dipl. -Ing. degree (magna
project 99E160. cum laude) from the Computer and Informatics Engineering
Department of the University of Patras, Greece, in 1997. He is
currently a Ph.D. candidate at the Telecommunications Labo-
ratory of the National Technical University of Athens, Greece.
References His research interests include distributed and mobile object
technologies, Intelligent Networks and telecommunications
service engineering. Mr. Perdikeas is a member of the Technical
[1] I. Venieris, H. Hussmann (Eds.), Intelligent Broadband Chamber of Greece.
Networks, Wiley, New York, 1998.
[2] I. Venieris, F. Zizza, T. Magedanz (Eds.), Object Oriented
Iakovos S. Venieris was born in Naxos, Greece, on 3 March
Software Technologies in Telecommunications, Wiley, 1965. He received the Dipl. -Ing. degree from the University of
New York, 2000. Patras, Patras, Greece in 1988, and the Ph.D. degree from the
[3] T. Magedanz, R. Popescu-Zeletin, Intelligent Networks ± National Technical University of Athens (NTUA), Athens,
Basic Technology Standards and Evolution, International Greece, in 1990, all in electrical and computer engineering. He is
currently an Associate Professor in the Electrical and Computer
Thomson Computer Press, London, 1996. Engineering Department of NTUA. His research interests are
[4] C. Gbaguidi et al., Integration of Internet and telecommu- in the ®elds of B-ISDN, intelligent networks, Internet, distrib-
nications an architecture for hybrid services, IEEE Journal uted processing, service and network control, and performance
on Selected Areas in Communications 17 (9) 1563±1579. evaluation. He has several publications in the above areas. Dr.
Venieris has received several national and international awards
[5] F. Chatzipapadopoulos et al., Harmonised Internet and for academic achievement. He has been exposed to standardi-
PSTN service provisioning, in: Special Issue on Mobile sation body work and has contributed to ETSI and ITU-T. He
Agents for Telecommunication Applications, Computer has participated in several European Union and national pro-
Communications Journal 23 (8) (2000) 731±739. jects. He is a member of IEEE and the Technical Chamber of
Greece.
[6] F. Chatzipapadopoulos et al., Mobile agent and CORBA
technologies in the broadband intelligent network, IEEE
Communications Magazine 38 (6) (2000) 116±124.

You might also like