Challenges in Multimedia Networking
Challenges in Multimedia Networking
A Research Agenda for Multimedia Networking
Mobiware: Mobile Middleware for Wireless ATM Networks
QOS Architectures
End-to-End QOS Management for Adaptive Video Flows
Design of a QOS Controlled ATM Based Communications
System in Chorus
The field of multimedia networks has literally exploded in the last couple of years. This can be seen not only from the increasing number of conferences and workshops organized in the past two years but also from the mergers and acquisitions fever in the meda, telephone, cable TV and publishing industries. Given the enormous scope of multimedia networking and the many changes it undergoes, it would be a garguantuan task to present an all encompassing picture of the challenges that lie ahead. Instead, we will limit ourselves here to a scenario of multimedia networks with Asynchronous Transfer Mode (ATM) based transport, such as already seen in broadband ATM LANS that are being deployed today. What are the main challenges that lie ahead for realizing these networks ? We believe that the following prerequisites will determine them (i) seamless connetion management between the network and multimedia devices (ii) multimedia abstractions with QOS guarantees and (iii) the integration of service, traffic control and network management. We will focus on each of these issues in the following paragraphs.
Connection Management (Binding) for Multimedia Networks. There is a realization in the IT industry that building open, programmable multimedia networks increases its capabilities and leads to innovation. As an example, the field of network management is often cited for making tremendous progress towards interoperability by applying a set of widely accepted standards to an open architecture. This is not yet the case in network control and connection management. The main reson for this is because current solutions to the problem tend to focus on low level issues using traditional tools and techniques that have evolved from the days of telephony. In comparison, the modern multimedia network is at least several orders of magnitude more complex than the simple telephone and we question the effectivenes of these traditional approaches to the problem at hand. We advocate a solution to the problem based on a paradigm that exploits the main capabilities and power of abstraction that distributed systems and object-oriented programming offer. The binding architecure advocated here is open : all the network entities that participate in the binding processes are modeled as objects with well defined interfaces. Communication among these objects is supported by the Object Management Group (OMG)'s Common Object Request Broker Architecture (CORBA).
Modeling Multimedia Networking Abstractions with QOS Guarantees. Abstractions with QOS guaranttes are well advanced in the broadband networking area. We propose a translation for them into the multimedia arena and thereby come up with a uniform virtual resource characterization for QOS. We do so by noting that current workstation architectures have inherent bottlenecks : I/O bottlenecks resulting from slow peripherals that were designed for small volumes of data; processing bottlenecks that are due to master/slave relationship between CPU and periperhal units; and finally performance bottlenecks due to excessive emphasis on system throughput without QOS considerations (e.g. delay bounds). In order to address these problems, a multiprocessor architecture for multimedia workstations seems appropriate. By employing a number of separate processors for various media, audio and video streams are supported by the first processor, the usual data or best effort applications are supported by a second processor and finally storage by a third processor. By isolating the execution of various multimedia applications to specialized processors, a characterization of the Audio Video Unit (AVU), Main Processing Unit (MPU) and Storage Unit (SU) as a device abstraction that guarantees QOS becomes possible. In this perspective, the concept of schedulable regions can be applied in the domain of multimedia devices.
Integration of Resource Control and Management. Network management architectures are typically designed around the basic manager agent model. Management information is stored in a Management Information Base (MIB) and is accessible remotely for by management applications for management tasks. When setting up connections that guarantee QOS, characterization of resources is needed for the purposes of defining optimal policies for allocating resources. Since integration between connection management and resource control is essential to the functioning of a network with QOS guarantees it is natural to consider an open architecture for the resource control as well. One obvious way to do this is to define the set of objects and interfaces that participate in the resource control task. As in the case of connection management, exchange of information among these objects will be supported by CORBA. This approach allows us to achieve an integration among connection management, resource control. We also believe that there is sufficient overlap in representations between objects defined using the network management GDMO and those defined using CORBA-based IDL to allow integration in our architecture with network management.
A full version of this work appears in [1].
A research agenda for multimedia networking is presented. The discussion centers around an extended reference model (XRM) that includes networking and multimedia computing devices. The XRM explicitly models the broadband and media processors, the multimedia network and the services and applications network. The interface between these models is given by quality of service and service abstractions. The reference model helps clarifying what a multimedia network is, how it differs from a broadband network, and how it differs from service and application networking. Research topics for future investigations are discussed throughout.
The Extended Reference Model (XRM). In parlance of network architectures, the figure below is an abstract representation of the Extended Reference Model. The XRM models the communications architecture of networking and multimedia computing platforms. The foundations for the operability of multimedia computing and networking devices is the same. Both classes of devices can be modeled as producers, consumers and processors of media. The only difference appears to be in the overall goal that a group of devices is set to achieve in the network or the multimedia platform.
The restriction of the XRM to networking is called the Integrated Reference Model (IRM). The IRM incorporates monitoring and real-time control, management, communication, and data abstraction primitives that are organized into five planes: the network and service management or N-plane, the resource control or M-plane, the data abstraction and management or D-plane, the connection management and binding or C-plane and the user information transport or U-plane (see figure below).
The restriction of the XRM to the multimedia computing platform has a similar functionality as the IRM. The N-plane includes system management functionality, and the M-plane includes process scheduling, memory management, routing (when applicable), admission control and flow control. The D-plane also contains objects modeling multimedia devices, the C-plane binding functionality, and the U-plane transport of user information within the Customer Premises Equipment.
Figure: The Extended Reference Model (XRM).
The Structure of the XRM. In order to make the structure of the XRM more transparent, we identify three (sub)models within the XRM. These models are defined by the functionality commonly associated with the broadband network and media processors, the multimedia network, and the applications and services network. We call these the R-, the G- and the B-models, or RGB for short. As the three colors that make out the spectrum of light (red, green and blue), these three models make out the XRM. The R-, G-, and B- models are shown in the figure below
.The interface between these models is defined by a set of well defined abstractions or services. QOS abstractions are provided by the broadband network and media proces- sors to the multimedia network, whereas service abstractions are defined at the interface between the multimedia network and the service and applications network.
Figure: RGB model.
The Functionality of the Broadband Network and Media Processors (R-Model). As already mentioned, the R-model, shown in the figure above, represents the functionality of the broadband network and the media processors. Overall, the broadband network and the media processors provide a service to the multimedia network in terms of QOS abstractions.
The D-plane abstracts the main network components, i.e., switches, multiplexers and media processors as a global distributed memory. Specifically, communication links are modeled as FIFO memory and, switches and processors as random access memory. These and higher level abstractions thereof are entities that constitute the Management Information Base (MIB).
The N-plane functionality is one of network and system management and comprises monitoring and control of individual states. These states might correspond to the status of a link, the temperature of a interface card, etc. and are enveloped as objects residing in the MIB. A client/server interaction is the basis of the N-plane manager/agent model for monitoring the controlling network elements.
The M-plane models the resource control tasks. For example, at the switch or multiplexer level the main resource functionality is in terms of buffer management and link scheduling. At the CPE level these same tasks appear as operating system scheduling and memory management. Flow control is another important resource control mechanism that acts on the cell level.
The C-plane supports exchange of state information among distributed buffer management and link scheduling entities. Exchange of state information is required in cooperative distributed scheduling. Such mechanisms can lead to a larger networking capacity under QOS constraints.
The U-plane defines cell level adaptation protocols for segmentation and reassembly as well as reliability checks. This include the ATM Layer and the ATM Adaptation Layer functions.
The Functionality of the Multimedia Network (G-Model). The functionality of the multimedia network is shown in the figure above. Again, this functionality is divided among the five planes of the XRM. The multimedia network provides to the services and applications network a set of service abstractions. As we already men- tioned in the previous section, these abstractions are realized by the multimedia network based on the QOS abstractions provided by the broadband network and media processors. Note here the similarity of our model with the OSI service model. The "layer" in our case is the multimedia network that provides services to the layer above (the services and applications network) based on services provided by the layer below (the broadband and media processors).
The D-plane consists of a Binding Interface Base (the BIB), a distributed repository containing information about entities that might participate in a binding process. Services are defined as a set of interconnected objects.
The binding architecture sets out to open and enlarge the applicability of proprietary signalling protocols. Currently, ATM LANs provide proprietary software in support of Q.93b or other closely related signalling protocols. This means that there is no interface for third party software providers for accessing network resources. There is a need to devise a methodology for an open network access to broadband networking and media processor resources with the aim of achieving a rapid and flexible service creation, deployment and management strategy.
Monitoring the behavior of distributed systems (such as Corba or the binding architecture) is a key requirement for the N-plane. This plane also provides for component management such as high speed manageable host interfaces.
The functionality of the U-plane includes the support of a number of media stream protocols such a native ATM stack and other real-time protocols. These protocols can coexist with a number of already widely used data protocols that offer best effort service (such as TCP).
The M-plane offers orchestration as well as other resource allocation mechanisms. The key resource control algorithms are routing and admission control.
Finally, the C-plane supports stream control as well as connection management and, more generally, binding algorithms. Stream control refers to protocols required for remote control of multimedia devices such as tape drives, multimedia on demand systems, etc. Both unicast as well as multicast connections management algorithms belong to the C-plane.
The Functionality of the Applications and Services Network (the B-Model). The functionality of the applications and services network is shown in the figure above.
The management of services is a functionality of the N-plane. Here we identify management support for access, security, configuration, billing and auditing services, among others. Service admission control is defined both in the N- and in the M-plane. How to control services and negotiate networking and computational resources is a key requirement of the M-plane. The D-plane is an object repository of services such as multimedia mail, computer supported cooperative work, video on demand, parallel virtual machines, etc. These are based on a service model whose structure will be discussed elsewhere. Navigation and service creation tools are the object of the C-plane. Finally, application protocols belong to the functionality of the U-plane.
Open issues. The description of the XRM and its subdivision into three functional reference models has already implicitly suggested a number of open issues in multimedia networking. The three key issues seem to be: (i) defining the interface between the multimedia network and, the broadband network and media processors, (ii) designing and implementing a QOS architecture for the multimedia network and, (iii) designing programming tools for supporting the services at the interface between the multimedia network and, the applications and services network.
Furthermore, scaling and operability are two major issues that will need to be dealt with extensively. Scaling refers to both time and space, and represents the ability to deal with more then ten orders of magnitude in time (from ns to minutes) and possibly global distribution of the network. Operability refers to the overall ability to manage and control; it implies the design of a system that gathers traffic statistics and responds rapidly to dynamically varying traffic loads, network status and fault conditions, and simultaneously provides different grades of service guarantees to different traffic types based on their fidelity requirements.
A full version of this work appears in [3].
The main Quality of Service (QOS) challenge in a combined wireline/wireless ATM network derives from the combination of providing multimedia services over multi-rate connections with mobility. A connection (aka flow) with certain capacity reserved at a particular cell may have to be re-routed to another cell when the mobile device changes its location. The new path to the desired location may not have the original required capacity. Therefore, re-negotiation of resources allocated to the connection is needed. At the same time, the flow (e.g., audio or video) should be transported and presented `seamlessly' to the mobile device with a smooth change of perceptual quality. In this work we describe QOS-aware mobile middle ware called mobiware which takes end-to-end programmability for QOS controlled mobility and the deliv ery of scalable multimedia flows over wireless ATM networks (WATM) as its primary design goal. We use the term "controlled QOS" to distinguish it from hard QOS guarantees offered by fixed ATM networks. Implicit in the term is the notion that mobile flows can be transported and represented as multi-resolution scalable flows at the mobile device.
Recent years have witnessed a tremendous growth in the use of wireless communications in business, consumer and military applications. The number of wireless services and subscribers has expanded with systems for mobile analog and digital cellular telephony, radio paging, and cordless telephony becoming widespread. Next generation wireless systems will provide enhanced communication services such as high resolution digital video and full multimedia communications. Recently there has been considerable interest in the emergence of wireless ATM (WATM) [1] as a possible candidate technology for next gener ation mobile multimedia communication systems [1] [2] [3] [4]. As multimedia applications migrate to mobile devices, wireless extensions to existing ATM networks are required to support the seamless deliv ery of voice, video and data with high quality. In this context WATM is intended to be a direct extension of the existing fixed/wireline broadband ATM network with uniformity of end-to-end QOS guarantees [1].
Figure:mobiware: mobile multimedia middlewar.
The development of next generation mobile multimedia communications systems presents a number of technical challenges which are thus far unresolved [5]. These challenges which are mainly due to large- scale mobility requirements, limited radio resources and fluctuating network conditions fundamentally impact our ability to deliver multimedia flows over mobile and QOS fluctuating networks. First, it is essential that QOS assurances be given for the transfer of audio and video flows to mobile devices as they roam between cells in cellular systems. Second, future mobile communications systems must be able to provide dynamic re-routing of a set of multimedia flows associated with a mobile device from one base station to another in a timely manner, without significantly interrupting the flows in progress and with a smooth change in the delivered quality. Third, existing multimedia transport systems [6] are ineffective when operating in environments where widespread mobility and changing network characteristics are dominant. New transport systems need to be designed that can operate in the face of QOS-varying channels and rapid device mobility. These protocols are yet to be designed and implemented.
Although researchers have addressed the isolated areas of architecture and algorithms for mobility, trans port, and the provision of end-to-end QOS provision [7] in mobile networks, little attention has been directed toward the development of a QOS programmable mobile middleware platform which supports the seamless transport and delivery of scalable multimedia flows over mobile networks with end-to-end QOS assurances.
To address this deficiency, we are developing QoS-aware mobile middleware called mobiware which builds QOS programmability, transportability and mobility into mobile ATM networks. The mobile middle ware platform which is being implemented in the fixed and mobile devices, base stations, and mobile- aware and standard ATM switches is the realization of: a mobile-aware QOS architecture and a family of adaptive algorithms for QOS-aware, mobile-aware and content-aware transport of scalable mobile flow [8] over WATM networks. As illustrated in the Figure, mobiware operates at the intersection of a number of sys tem facets offering end-to-end QoS control, mobility support and the transport of scalable flows. mobi ware supports a high degree of middleware programability through the use of xbind which is based on CORBA technology.
A full version of this work appears in [7].
References
[1] Raychaudhuri, D., (NEC USA), Dellaverson, L., (Motorola), Umehira, M., (NTT Wireless Systems), Mikkonen, J., (Nokia Mobile Phones), Phipps, T., (Symbionics), Porter, J., (Olivetti Research), Lind, C., (Telia Research) and Suzuki, H., (NEC C & C Research), ``Scope and Work Plan for Proposed Wireless ATM Working Group,'' ATM Forum Technical Committee, ATM Forum/96-0530/PLEN, April 1996.
[2] Raychaudhuri, D and Wilson, N., ``ATM-Based Transport Architecture for Multiservice Wireless Personal Com munications Networks,'' IEEE Journal on Selected Areas in Communications, Vol. 12., No. 8. October, 1994.
[3] Porter, J., Hopper, A., Gilmurray, D., Mason, O., Naylon, J., and A. Jones, ``The ORL Radio ATM System, Architecture and Implementation,'' Technical Report, ORL Ltd, Cambridge, UK, January, 1996.
[4] First Wireless ATM Networking Workshop: State-of-the-art and Beyond, Columbia University, New York, NY, June 14, 1996. http://comet.ctr.columbia.edu/activities/opensig/activities/watm.html
[5] Aurrecoechea, C., Campbell, A.T. and L. Hauw, ``A Survey of QOS Architectures,'' Multimedia Systems Journal, Special Issue on QOS Architecture, 1996, (to appear)
[7] Campbell A. T.,
``Incorporating QoS Adaptation into Multimedia Transport Systems,''
IEEE Network, 1996 (to appear)
[8] Campbell A..T., ``Towards End-to-End Programmability for QOS Controlled
Mobility in ATM Networks and their Wireless Extensions,''
Proc. 3rd International Workshop on Mobile Multimedia Communications
(MoMuC-3),
Princeton, Sept. 25-27, 1996, and Wireless ATM Workshop, Espoo, Finland,
Sept. 2-3, 1996, (invited presentation)
[9] Balachandran, A., and A. T. Campbell,
``Java Agents: An Approach toward Media Scaling For Wireless Communications,''
Technical Report, Center for Telecommunications Research, Columbia University, October, 1996.
Over the past several years there has been a considerable amount of research within the field of quality of service (QoS) support for distributed multimedia systems. To date, most of the work has been within the context of individual architectural layers such as the distributed system platform, operating system, transport subsystem and network layers. Much less progress has been made in addressing the issue of overall end-to-end support for multimedia communications. In recognition of this, a number of research teams have proposed the development of QoS architectures which incorporate quality of service configurable interfaces and quality of service driven control and management mechanisms across all architectural layers. This work examines the state-of-the-art in the development of QoS architectures. The approach taken is to present QoS terminology and a generalised QoS framework for understanding and discussing quality of service in the context of distributed multimedia systems. Following this, we evaluate a number of QoS architectures that have emerged in the literature.
Meeting quality of service (QoS) guarantees in distributed multimedia systems is fundamentally an end-to-end issue, that is, from application-to-application. Consider, for example, the remote playout of a sequence of audio and video: in the distributed system platform, quality of service assurances should apply to the complete flow of media from the remote server across the network to the point/s of delivery. As illustrated in the Figure, this generally requires end-to-end admission testing and resource reservation in the first instance, followed by careful co-ordination of disk and thread scheduling in the end-system, packet/cell scheduling and flow control in the network and, finally, active monitoring and maintenance of the delivered quality of service. A key observation is that for applications relying on the transfer of multimedia and, in particula, continuous media flows, it is essential that quality of service is configurable, predictable and maintainable system-wide, including the end-system devices, communications subsystem and networks. Furthermore, it is also important that all end-to-end elements of distributed systems architecture work in unison to achieve the desired application level behaviour.
To date, most of the developments in the area of quality of service support have occurred in the context of individual architectural components. Much less progress has been made in addressing the issue of an overall QoS architecture for multimedia communications. There has been, however, considerable progress in the separate areas of distributed systems platforms, operating systems, transport systems and multimedia networking [46-66] support for quality of service. In end-systems, most of the progress has been made in the areas of scheduling, flow synchronisation and transport support. In networks, research has focused on providing suitable traffic models [2] and service discipline, as well as appropriate admission control and resource reservation protocols. Many current network architectures, however, address quality of service from a providers point of view and analyze network performance, failing to comprehensively address the quality needs of applications. Until recently there has been little work on quality of service support in distributed systems platforms. What work there is has been mainly carried out in the context of the Open Distributed Processing.
The current state of QoS support in architectural frameworks can be summarized as follows:
In recognition of the above limitations, a number of research teams have proposed systems architectural approaches to QoS support. In this work these are referred to as QoS architectures [67-90]. The intention of QoS architecture research is to define a set of quality of service configurable interfaces that formalize quality of service in the end-system and network, providing a framework for the integration of quality of service control and management mechanisms.
In this work we present a generalized QoS framework and terminology for distributed multimedia applications operating over multimedia networks with QoS guarantees. The generalized QoS framework is based on a set of principles that govern the behavior of QoS architectures. In addition, we evaluate a number of QoS architectures found in the literature that have been developed by the telecommunications, computer communications and standards communities and then present a short qualitative comparison and discussion.
A full version of the work can be found in [4].
References
[1] Keshav, S., ``Report on the Workshop on Quality of Service Issues in High
Speed Networks,'' ACM Computer Communications Review, Vol 22, No 1,
pp 6-15, January, 1993.
[2] Kurose, J.F., ``Open Issues and Challenges in Providing Quality of Service
Guarantees in High Speed Networks,'' ACM Computer Communications Review,
Vol 23, No 1, pp 6-15, January 1993
[3] Aurrecoechea, C., Campbell A. T. and L Hauw,
``A Survey of QoS Architectures,''
Multimedia Systems Journal, 1997 (to appear)
[4] Aurrecoechea, C., Campbell A. T. and L Hauw,
``Architectural Perspective on QoS Management in Distributed Multimedia Systems,''
Proc. 2nd Int'l Workshop on Protocols for Multimedia Systems. Salzburg, Austria, Oct. 1995.
Andrew Campbell, Alexandros Eleftheriadis and Cristina Aurrecoechea
Distributed audio and video applications need to adapt to fluctuations in
delivered quality of service (QOS). By trading off temporal and spatial quality
to available bandwidth, or manipulating the playout time of continuous media
in response to variation in delay, audio and video flows can be made to adapt
to fluctuating QOS with minimal perceptual distortion. In this work we
introduce dynamic QOS management (DQM) for the control and management
of multi-layer coded flows operating in heterogeneous multimedia networking
environments. Two key techniques are proposed: i) an end-to-end dynamic rate
shaping scheme which adapts the rate of MPEG-coded flows to the
available network resources while minimising the distortion observed at the
receiver; and ii) an adaptive network service, which offers hard guarantees to
the base layer of multi-layer coded flows, and fairness guarantees to the
enhancement layers based on a bandwidth allocation technique called weighted
fair sharing. We also discuss a number of types of media scaling objects which
are used to manage and control end-to-end QOS. These include QOS filters
which manipulate multi-layer coded flows as they progress through the
communications system, QOS adaptors which scale flows at end-systems based
on the flow's measured performance and user supplied QOS scaling policy, and
QOS groups which provide baseline QOS for multicast flows.
In this work we introduce a set of media scaling objects used to manipulate
flows as they progress through the communications system. These comprise
QOS adaptors, QOS filters and QOS groups: QOS adaptors are used in
conjunction with flow monitoring function to
ensure that the user and provider QOS specified in the service contract are
actually maintained. In this role QOS adaptors are seen as quality of service
arbiters between the user and network. QOS adaptors scale flows at the end-systems
based on a user supplied QOS scaling policy
and the measured performance of on-going flows.
QOS filters manipulate multi-layer coded flows at the end-systems
and as they progress through the network. We describe three distinct styles of
QOS filters:
In [2] we designed a service contract based API which formalised the end-to-end
QOS requirements of the user and the potential degree of service
commitment of the provider. An important aspect of our API is that it shields
the application from the complexity of QOS management and control [2]. In
this paper, we detail extensions to the flow specification, QOS commitment
and QOS scaling clauses of the service contract required to accommodate
adaptive multi-layer flows. The API presented here is not complete in that
there are no primitives given for establishing and renegotiating connections or
for manipulating QOS groups. Full details of these aspects are given in [2].
Multi--layered flows are characterised by three sub--signals in the flowSpec flow
specification: a base layer (BL) and up to two enhancement layers (E1 and E2).
Each layer is represented by a frame size and subjective or perceptive QOS as
illustrated in [3]. Based on these characteristics, the MPEG--2 coder
determines approximate bit rate for each sub--layer. In the case of MPEG--2's
hybrid scalability [2], BL would represent the main profile bit rate requirement
(e.g. 0.32 Mbps) for basic quality, E1 would represent the spatial scalability
mode bit rate requirement (e.g. 0.83 Mbps) for enhancement, and E2 would
represent the SNR scalability mode bit rate requirement (e.g. 1.85 Mbps) for
further enhancement. The remaining flow specification performance
parameters for jitter, delay and loss are assumed to be common across the all
sub--signals (i.e. a single layer of a multi--layer video flow). The QOS
commitment field has been extended to offer an adaptive network service that
specifically caters for the needs of scalable audio and video flows in
heterogeneous networking environments (see adaptive network service
section).
The QoSscalingPolicy field of the flowSpec characterises the degree of
adaptation that a flow can tolerate and still achieve meaningful QOS. The
scaling policy consists of clauses that cover adaptation modes, QOS filter
styles, and event/action pairings for QOS management purposes. Two types of
adaptation mode are supported: continuous mode, for applications that can
exploit any availability of bandwidth above the base layer; and discrete mode
for applications which can only accept discrete improvement in bandwidth
based on a full enhancement (viz. E1, E2). The QOS scaling policy provides
user--selectable QOS adaptation and QOS filtering. While receivers select filter
styles to match their capability to consume media at the receiver (from the set
of temporal filters), senders select filter styles to shape flows in response to the
availability of network resources such as bandwidth and delay (from the set of
shaping filters). Network oriented filters (i.e. selection filters) can be chosen by
either senders or receivers. In addition, senders and receivers can both select
periodic performance notifications including available bandwidth, measured
delay, jitter and losses for on--going flows. The signal fields in the scaling
policy allow the user to specify the interval over which a QOS parameter is to
be monitored and the user informed. Multiple signals can be selected
depending on application needs.
A full version of this work appears in [1].
References
[1]
A. Campbell, A. Eleftheriadis and C. Aurrecoechea, ``End-to-End QoS Management of
Adaptive Flows,'' IEEE Symposium of Multimedia Communications and Video Coding,
New York, October 1995.
[2]
A. Campbell, D. Hutchison and C. Aurrecoechea, ``Dynamic QoS Management for Scalable
Video Flows,'' Proc. Fifth International Workshop on Network and Operating System
Support for Digital Audio and Video, Durham, New Hampshire, 1995.
[3]
A. Campbell, A. Eleftheriadis and C. Aurrecoechea, ``Meeting End-to-End QoS Challenges for
Scalable Flows in Heterogeneous Multimedia Environments,'' Proc. 6th IFIP International
Conference on High Performance Networking}, Palma, Spain, September 1995.
G. Blair, A. Campbell, G. Coulson, D. Hutchison, M. Papathomas and P. Robin
In the recent past, significant research has been carried out into communication
s systems
for distributed real-time and multimedia applications. A surprisingly small amount of this
work, however, has considered the issues that arise when high-speed communications systems
are interfaced to conventional workstations running standard multiprogrammed operating
systems such as UNIX. Rather, the research has tended to focus on network issues
or has made specific assumptions about the end points of multimedia and real-time
communication. Some researchers have considered abstract transport protocol specifications or
service interfaces. Others have assumed specialised end-systems such as
CODECS or multimedia enhancement units. Still others have
considered computers running specialised real-time operating systems unable to support
conventional applications or conventional modes of operation (e.g. such systems
typically preclude dynamic process creation).
The research is aimed at providing system software support for
distributed real-time and multimedia applications in an environment of standard
workstations
and high-speed networks. Our specific aims are as follows:-
Chorus has several desirable real-time features and has been widely used for embedded
real-time applications. Real-time features include pre-emptive scheduling, page
locking,
timeouts on system calls, and efficient interrupt handling. Unfortunately, Chorus real-time
support is not fully adequate for the requirements of distributed real-time and
multimedia
applications, principally because there is no support for QOS specification and
resourcereservation:-
The advantage of lightweight threads and user level scheduling is that context s
witch
overhead is minimal. The drawback of user level scheduling is that, by definition, it cannot
ensure that CPU resources are fairly shared across multiple actors; this is the
role of kernel
level scheduling. The split level architecture combines the benefits of both user level and kernel
level scheduling by maintaining the following invariants:-
i) each ULS always runs its most urgent lightweight thread,
Note that the notion of urgency is dependent on the scheduling policy used (e.g.
it would
be deadline for EDF scheduling and priority for rate monotonic scheduling).
The necessary information exchange between the KLS and the ULSs is accomplished
via a
combination of shared memory and upcalls from the kernel. The per-VP shared
memory areas contains the urgency of the most urgent lightweight thread known to
this VP.
This is read by the KLS on each kernel level rescheduling operation to determine
the next VP to schedule.
A passive shared memory area between the KLS and the ULSs is not sufficient. It
is also
necessary for the KLS to be able to actively interrupt VPs to inform them of the
occurrence of
real-time events in a timely fashion. Such events include timer expirations used
to implement
pre-emption in user level scheduling, and data arrivals from local kernel device
s or from the
network device. We use a software interrupt mechanism for the notification of such events.
Software interrupts are always targeted at application actors but can be initiated either by kernel
components (e.g. the KLS) or by library code in other application actors.
The last major component of the scheduling architecture is the use of non-blocking system
calls to avoid potential violations of the scheduling invariants. With traditional
blocking system calls such violations can occur when a lightweight thread performs a blocking
system call which blocks its underlying VP. This disallows another lightweight thread in the
same actor from executing while the blocking call is extant - even if it is the
globally most
urgent thread. Non blocking system calls avoid this problem by returning immedia
tely and thus
allow the calling VP to run another lightweight thread while the system call is
being serviced.
The result of the call is eventually notified by a software interrupt as discuss
ed above.
The standard Chorus communications stack was designed for the support of connectionless
datagram services and uses retransmission strategies to enhance reliability. In
contrast, our
communications architecture is intended to support connection oriented
communications over ATM networks with QOS support and configurable error control.
Because of these distinct design goals, we have initially designed our stack to
operate entirely
separately from the existing Chorus facilities although we intend in the future
to integrate the
functionality of the two stacks in a unified architecture.
A full version of this work appears in [1].
References
[1]
G. Coulson, A. Campbell, P. Robin, G. Blair, and M. Papathomas,
``Design of a QoS Controlled ATM Based Communications System in Chorus,''
IEEE Journal of Selected Areas in Communications (JSAC),
Special Issue on ATM LANs Implementation and Experiences with an Emerging Technology, 1995.
Cristina Aurrecoechea, Andrew T. Campbell and Linda Hauw
QOS Architectures
End-to-End QOS Management for Adaptive Video Flows
Before potential senders and receivers can communicate they must first join a
QOS group. The concept of a QOS group is used to associate a baseline QOS
capability to a particular flow. All sub-signals of a multi-layer stream can be
mapped into a single flow and multicast to multiple receivers. Then, each
receiver can select to take either the complete signal advertised by the QOS
group or a partial signal based on resource availability. Alternatively each subsignal
signal can be associated with a distinct QOS group. In this case, receivers
tune into different QOS groups (using signal selection) to build up the overall
signal. Both methods are supported in DQM. Receivers and senders interact
with QOS groups to determine what the baseline service is, and tailor their
capability to consume the signal by selecting filter styles and specifying the
degree of adaptability sustainable; that is whether flows are discrete or
continuous adaptive.
Design of a QOS Controlled ATM Based Communications System in Chorus
Our approach is to use a micro-kernel operating system, specifically Chorus, as
a system
software support layer for both UNIX and real-time applications. A standard UNIX
SVR4
personality included with Chorus is used to support UNIX applications. Our exten
sions to
Chorus described in this paper are used to support real-time applications. The present paper
focuses on the resource management strategies used in our Chorus extensions. The
three major
resource classes considered are CPU cycles, network resources and physical memory. Note
that in this paper we focus on end-system related communications issues rather than internet or
network resource management issues (although we do cover resource allocation in
the ATM
network environment).
To remedy its current deficiencies for QOS specification and real-time application support,
we have extended the Chorus system call API with new low level calls and abstractions. The
new abstractions, provided in both the kernel and a user level library are described below.
The scheduling architecture exploits the concept of lightweight threads which are supported
in a user level library and multiplexed on top of a single Chorus kernel thread
which, in this
context, we refer to as a virtual processor (VP). The scheduling architecture is
a split level
structure consisting of a single kernel scheduler (KLS) to schedule VPs, and
per-actor user level schedulers (ULSs) to schedule lightweight threads on those
VPs.
ii) the KLS always runs the VP supporting the globally most urgent lightweig
ht thread.