Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from A-E

Distributed Multimedia Systems - Overview, Key Technologies and Solutions in the Design of a DMMS

service issues caching qos

Bharadwaj Veeravalii
National University of Singapore, Singapore

Definition: Distributed multimedia systems consist of multimedia databases, proxy and information servers, and clients, and are intended to for the distribution of multimedia content over the networks.

In this article, we identify most imperative issues in the design of DMMS architecture. The article is by no means is a survey of DMMS; however it is expected to bring out the key issues and challenges in this domain. We present comprehensive discussions pointing to several key papers published in the literature and highlight existing solutions for small and large-scale DMMS.

Overview

Distributed Multimedia System (DMMS) architecture with all its essential components (Multimedia Databases (MMD), Proxy servers, information services, etc.) is shown in Figure 1.

In a large scale network infrastructure, it is wiser if the control is distributed in the sense that service providers (SPs) choose vantage sites to exercise control to regulate the traffic which in a way assures a respectable quality of service (QoS). This is facilitated via agent-driven approaches. Thus a central dogma in modern days is in adopting agent-driven support within a DMMS to handle overwhelming client population on the network. A

decade ago, networked multimedia systems were capable of supporting mostly devices like Personal computers and/or a small LAN set-up. However, with the advent of modern day wireless technology, devices such as mobile-technology enabled laptops, handheld devices such as palm-tops, PDAs, etc also fall under active interactive devices. This means that in the volume of traffic ranges from simple short media clips, images, and text messages to long duration media data, which is a multi-fold increase. Further, when compared to service architectures conceived from late 80’s to mid-90’s, modern day services need to account radically different issues in addition to the existing issues in the design of a DMMS architecture. To appreciate this point, one can quickly imagine the issues related to a mobile technology playing crucial roles such as ensuring continuous network connectivity, graceful degradation of service quality under unavoidable circumstances, replication of media data 1 and maintaining consistency for editable data, if any, to quote a few. In addition to such media-rendering service facilities, the purview of modern day DMMS extends to entertainment in the form of games and casinos on networks.

From client’ perspective the demands are very simple – design of DMMS must be completely flexible. One would expect different kinds of service facilities ranging from simple pay-per-view shows to an interactive mode of movie viewing for Video/Movie-On-Demand (VoD/MoD). In each of the above expectations, design of DMMS must account a wide variety of issues. Finally, by resources, one would mean the following: adequate number of copies of movies at the multimedia databases(MMDs), memory capacity at intermediate nodes to buffer data on its transit to destination, adequate bandwidth of the network to support media streams (inclusive of both time-continuous and non-continuous data), and any plug-and-play facilitating software modules. The last component is meaningful under agent-driven systems wherein need for an agent migration and number of agents to be generated at any instance are critical issues.

Key Technologies and Solutions in the Design of a DMMS

In the previous section, we have highlighted some differences between conventional DMMS architectures and expectations of modern day DMMS architecture in providing a wide range of services. In a nut-shell, owing to high-bandwidth availability, several applications become plausible for users that will enforce a continuous work pressure on the media servers on the network. Thus managing resources is indeed a key challenge. In Table 1, we shall present a list of issues which may be considered as either primary or secondary depending on the context and module under consideration. This list, by no means exhaustive, however attempts to capture significant issues that are in practice.

Major Service Categories: Video-On-Demand (VoD) versus Video-On-Reservation(VoR): VoD and VoR are most commonly used services by network users. VoD is certainly an attractive technology in rendering digital video libraries, distance learning, electronic commerce, etc, as (i) users can request the service any time and from anywhere and, (ii) service allows users to have a complete control of the presentation. Despite continuous research efforts in rendering quality VoD service, the requirement on network resources, such as server bandwidth, cache space, number of copies, etc., are still considered overwhelming owing to an exponential growth in client population. When such fully interactive service demands are met we say that the service is of type True-VoD. While small movie clips rendering, learning and video-conferencing kind of applications are almost now delivering a high-quality service, for long-duration movie viewing(say HOmins standard movie) with a presentation control still seems to exhibit annoying experiences for users. Article on long-duration movie retrievals can be found in . Contrary to this, when users reserve for a movie presentation in advance, VoD manifests in the form of VoR. Under VoR service, system is shown to utilize resources in an optimal manner as user viewing times are known in advance. VoD and VoR are completely orthogonal in their service style. Perhaps VoR is better suited for pay-per-view by SPs for digital TV subscribers. Another type of VoD service is called as Near VoD services (NVoD), and it distributes videos over multicast or broadcast channels to serve the requests which demand the same videos and arrive close in time. These technologies have been successful to provide video services in local area networks, say in hotels.

Media Streaming as a solution to handle large client pool : Fundamentals on media types and concepts pertaining to what media streaming is can be found in. A straightforward technology to realize streaming is to unicast streams is by allocating dedicated resources for each individual client session. As this is never a cost effective solution and non-scalable, IP-based multicast approaches were proposed. Techniques such as Batching, Patching, Merging, Periodical and Pyramid broadcasting, etc fall into this category. These can be found in . Several variants also exist in the literature to these fundamental schemes. Here, multiple requests share a video stream via the network; thereby the required server network I/O bandwidth can be reduced.

Various types of Media Distribution : There exist several practical approaches to media distribution, ranging from distributing small sized clips to long movies. The solution attempts include: proxy-based approach, parallel or clustered servers approach, and co-operative server scheduling. In proxy-based approach, a cluster of proxies are placed at vantage sites which can intercept the client requests or client requests can be redirected to these proxies for balancing the overall load and also to minimize access delays. However, in parallel or clustered servers approach, the system sends each request to all the servers and all the servers participate equally to serve the request. This mechanism is shown to enable to improve the system throughput and balance the load of each server. In the case of co-operative type, servers cooperate in a joint caching and scheduling of streams and the typical problem is in deciding when, where, and how long to cache a stream before it can be delivered to the clients and this caching on-the-fly is done for serving several requests demanding the same stream.

Caching in DMMS : Caching is one of the inevitable and powerful solution approaches that influence almost every performance metric under consideration. “Cache-or-not-to-Cache” is often a most common dilemma of the algorithms in place as the decision is based on a combined influence of several parameters. Caching allows nodes to quickly fetch the required documents without incurring annoying delays by circumventing the need to contact the original host. Caching can be at memory/disk level or at node level. Again, performance of a DMMS system can also be quantified in terms of local as well in global terms. In memory caching, the high-speed main memory is referred to as the cache of the relatively slow-speed disk, while in disk caching, a near-distance disk (e.g., a proxy) is used as the cache of the far-distance disk (e.g., in original server), or the disk is used as the cache of tertiary storage, e.g., CD, tape. For modern day applications, even acquiring a small amount of an intermediate storage seems a critical issue to account for. Day-to-day storage devices have capacities ranging from 18GBytes cache space (Seagate cheetah) to 250Mbytes (Toshiba Flash memories) and bandwidths in MB/sec in the range of 63.2 to 10, respectively. To improve the cache space of storage devices, Redundant Arrays of Inexpensive Disks (RAID) was proposed in to combine multiple small, inexpensive disk drives into an array of disk drives. This disk array can yield performance exceeding that of a single large disk drive related to space, speed and data protection. Apart from these technology oriented advantages, caching schemes such as Interval Caching, Block caching, Resource-based caching, Multi-policy integrated caching , etc, attempts to maximize the throughput of the system by cleverly caching streams of media either on-the-fly or on demand. Compilation in gives details on some of these techniques.

Miscellaneous Issues : Other issues at a higher level (system design level) that influence the performance critically include, multimedia databases (access and retrieval), object oriented development of such multimedia information systems, monetary cost optimization, movie placement at vantage sites. Some peripheral issues concerned include, building a distributed multimedia archive, an efficient search engine with or without agent-driven characteristics, an information repository (similar to yellow-page e-directory services), a flexible GUI to user and tracking user behavior. At the lower levels, issues concerned range from data organization, disk scheduling, and retrieving of data to memory management at node and system levels.

1 This is one way of maximizing the availability.

Some Key Performance Metrics and Useful Results

Three main key performance metrics that are akin to DMMS are discussed below.

Quality of Service (QoS) : Meeting a high-quality QoS demands is often a conflicting issue. From user’s perspective it is the timely delivery, full presentation control without any jitters and minimum access delays, while from SPs perspective optimal use of resources for maximizing the number of clients and meeting their demands. With Internet’s modern day best effort point-to-point services, quality can be assured. However, for multimedia traffic 2 QoS management is much more challenging. In the literature, there exist several models to capture statistical behavior and nature of multimedia traffic. These include, Markov modulated process models, Fractional Brownian motion model, Hybrid Bounding Interval Dependent (H-BIND) model, M/Pareto distribution model, Self-similar traffic model, and MPEG coded video traffic models, etc. The kind of parameters that a QoS management system must consider on the whole can be classified based on, (i) Network level QoS and, (ii) Application level QoS. Typical to the first category are transmission bit rates, delay, jitters, packet-loss, error rates, etc. IETF (Internet Engineering Task Force) has proposed several promising models such as Intserv (Integrated Service model) and Diffserv (Differentiated Service model), for supporting multimedia traffic over IP network. The essence of IntServ is to reserve resources (link bandwidth and buffer space) for each individual session so that the QoS can be guaranteed. Details on Diffserve and Intserv can be found in the above literature. A variety of QoS management architectures and technologies have been investigated to provide the desired QoS for the applications having divergent needs. The mechanism of QoS management includes connection acceptance, negotiation of the QoS, congestion control, traffic enforcement, and resource allocation. Several other possible QoS parameters are presented in. QoS aware techniques also exist and these belong to Network-layer and Application-layer multicasting categories.

Access Time : This is one of the key considerations for most of the services offered on networks. This is the time duration for which a subscribed user has to wait to avail the service. In VoD case, this is the difference in time between the instant when a user places a request and when the video is available for viewing at the client end. This is more of an issue that is usually from user’s perspective and even SPs may use this as their “service promise” in rendering high quality services. Several techniques are in place to minimize this access time. These include, pyramid broadcasting, multiple servers retrieval, careful placement of copies of movies at vantage sites for quick access, replication of copies of movies, etc, are a few to quote For large-scale networks this is an important metric to be considered by a SP.

Acceptance Ratio/Blocking Probability and Throughput : This metric is particularly meaningful when requests are for small clips and do not require significant amount of resources to consume. For instance, for VoD or pay-per-view kind of applications, as clients are primarily subscribed with the SP, it will not be meaningful if SP declines a client’s request. However, when requests arrive in a time-continuous fashion demanding shorter service requirements, depending on the current loading some requests may be dropped, which leads to these metrics. Also, quantifying the amount of throughput or equivalently the fraction of accepted or rejected requests usually considers a fairly large amount of observation over time by any algorithm/SP. These metrics work in conjunction with caching and cache replacement algorithms that are in place. Hence, most of the literature presenting cache performance also considers these metrics.

2 Constant bit-rate (CBR) or variable bit-rate (VBR), bursty are a few to quote.

Conclusions

In this article we presented an overview of modern day DMMS and summarized all the primary issues leading to the design of DMMS. The literature cited here is by no means exhaustive as DMMS is an ever challenging and growing domain and subsumes several issues in a complex way. A good design of DMMS must attempt to satisfy the common three-letter rule “F-A-S” (can also be referred to as performability ) – i.e., it must consider Flexibility (in terms of adapting to client’s needs and accommodating other applications), Availability of service, and Scalability (in terms of providing a quality service when client population and size of the network grows). As a final remark, a modern day DMMS also considers wireless networks as a connection medium for small scale multimedia applications. However, supporting continuous streaming and interactive multimedia applications is still a major challenge ahead for wireless systems.

Divorce His (1973)/Divorce Hers (1973) - Overview, Synopsis, Critique [next] [back] Distanced Leadership and Multimedia - INTRODUCTION, THE PROLIFERATION OF DISTANCED LEADERSHIP, CONCEPTUALIZATIONS OF “DISTANCE”, MULTIMEDIA AND DISTANCED LEADERSHIP, Effective Practices

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or