As Internet services grow in complexity, Internet Service Providers (ISPs) are finding out that the ad hoc methods that they have employed thus far to monitor and diagnose their services are not sufficient to provide acceptable service quality to their subscribers. In this paper, we demonstrate how service models can he used by ISPs to effectively manage their service offerings. A service model encapsulates a human expert's knowledge of a service, its components, and its interdependencies with other services. In addition, using ongoing measurements, a service model tracks the health of the different services and their components. By traversing a service model top-down, an operator can not only assess the overall health of a service, but also easily correlate the health of all the services and service components to determine the root cause of any problems that may occur. By minimizing the time and effort needed to diagnose problems, service models enable ISP operators to efficiently resolve problems that occur in an ISP environment. Since each ISP system is unique in many respects, unique service models have to be crafted for each of the services in every ISP system. Handcrafting customized service models requires enormous effort and time on the part of a human expert, a luxury that few ISPs can afford. In this paper, we describe a methodology for constructing customized service models for a target ISP system with minimal human intervention. This methodology relies on a service model creation engine that composes a custom service model for an ISP system using a predefined service model template specification and automatically discovered information about the target ISP system. We describe a prototype implementation of this methodology and present an example of a service model obtained from a real-world ISP system. The concepts described are applicable for the management of networks and services in enterprise systems as well.

No abstract available.

Motivated by the phenomenal growth of the Internet in recent years, a number of cable operators are in the process of upgrading their cable networks to offer data services to residential subscribers, providing them direct access to a variety of community content as well as to the Internet. Using cable modems that implement sophisticated modulation-demodulation circuitry, these services promise to offer a several hundredfold increase in access speeds to the home compared to conventional telephone modems. Initial experiences indicate that cable networks are susceptible to a variety of radio-frequency (RF) impairments that can result in significant packet loss during data communication. In the face of such losses, the transmission control protocol (TCP) that is predominantly used by data applications degrades dramatically in performance. Consequently, subscribers of broad-band data services may not perceive the projected hundredfold increase in performance. We analyze the performance of TCP under different network conditions using simulations and propose simple modifications that can offer up to threefold increase in performance in access networks that are prone to losses. These modifications require only minor changes to TCP implementations at the local network servers alone (and not at subscribers' PCs).

We define and formulate various policies for load management in distributed video servers. We propose a predictive placement policy that determines the degree of replication necessary for popular videos using a cost based optimization procedure based on a priori predictions of expected subscriber requests. For scheduling requests, we propose an adaptive scheduling policy that compares the relative utilization of resources in a video server to determine an assignment of requests to replicas. To optimize storage utilization, we also devise methods for dereplication of videos based on changes in their popularities and in server usage patterns. Performance evaluations indicate that a load management procedure which uses a judicious combination of the different policies performs best for most server configurations. Advances in storage technologies are making high performance video servers a reality. These video servers are being deployed over emerging broadband networks to deliver a variety of interactive, digital video services to thousands of residential subscribers. To meet the scalability requirements in such large deployments, distributed video server architectures are being considered (M. Buddhikot and G. Parulkar, 1995). We propose various methods for load management that are targeted at improving the cost effectiveness of distributed video servers.

Future advances in storage and networking technologies will make it feasible to build multimedia on-demand information servers capable of providing services similar to those of a neighborhood videotape rental store over metropolitan area networks. Such multimedia information servers must not only support retrieval of continuous media units (such as video frames and audio samples), but also preserve synchrony among playback of the different media components constituting a multimedia object. We develop techniques for supporting continuous and synchronous retrieval from multimedia servers. We present feedback techniques by which, during retrieval of multimedia objects from a multimedia server to mediaphones, the multimedia server uses lightweight messages called feedback units transmitted periodically back to it (by mediaphones) to detect impending discontinuities as well as asynchronies at mediaphones. The multimedia server then preventively readjusts media transmission so as to avoid either anomaly, and steers the mediaphones back to synchrony. Given the available buffer sizes at mediaphones and the maximum tolerable asynchrony, we present methods to determine the minimum rate at which feedback units must be transmitted so as to maintain both continuity and synchronization. These feedback techniques remain robust even in the presence of playback rate mismatches and network delay jitter, and their initial simulation for video-audio playback yields a feedback rate of one per 1,000 media units to keep the asynchrony within 250ms, showing that the overhead due to feedback transmission is very small. The constant rate feedback techniques developed in this article form the basis of a prototype on-demand information server being developed at the UCSD Multimedia Laboratory.

Future advances in networking and storage will make it feasible to build multimedia on-demand servers that provide services similar to those of a neighbourhood videotape rental store over metropolitan area networks such as B-ISDN. Such multimedia servers can support real-time retrieval of multimedia objects by users onto their ISDN videophones and audiophones for playback. The design of techniques and protocols for providing continuous and synchronous access to multimedia services constitutes the subject matter of this paper. In future integrated networks, mediaphones that possess bare minimum capability to playback media but which lack the sophistication to run elaborate time synchronization protocols, may be connected directly to the network. We present rate-based feedback strategies by which, during retrieval, a multimedia server uses light-weight messages called feedback units transmitted periodically to mediaphones, to accurately estimate the playback instants of media units. Using these estimates, the multimedia server detects impending playback discontinuities due to buffer overruns or starvations at mediaphones, and preventively readjusts media transmission so as to avoid either of these anomalies. Given the available buffer sizes at mediaphones, we present methods by which a multimedia server can determine the minimum rate at which feedback units must be transmitted to it by the mediaphones, so as to maintain continuity of media playback. In order to guarantee synchronous playback at mediaphones, we first propose a bounded buffering technique which uses buffering limitations, at the slave mediaphones to automatically enforce bounds on the asynchrony among mediaphones. Although simple and easy to implement, this technique may entail a large average asynchrony, in order to avoid which we propose a multiple feedback synchronization technique. We present initial performance comparisons of the effectiveness of these synchronization techniques. The techniques for intra-media continuity and inter-media synchronization presented in this paper from the basis of a prototype multimedia on-demand server being developed at the UCSD Multimedia Laboratory.

The authors have developed intermedia synchronization techniques for multimedia on-demand retrieval over integrated networks in the absence of global clocks. In these techniques, multimedia servers use lightweight messages called feedback units transmitted by media display sites (such as audiophones and videophones, generically referred to as mediaphones) to detect asynchronies among those sites. They present strategies by which the multimedia server can adaptively control the feedback transmission rate from that mediaphone, so as to minimize the associated overheads without permitting the asynchrony to exceed tolerable limits. They compare the performance of various resynchronization policies such as conservative, aggressive, and probabilistic. Performance evaluation of the feedback techniques indicates that their overheads are negligible; for a typical audio/video playback environment, the feedback frequency was about one in a hundred.>

The problem of media mixing that arises in teleconferencing applications such as teleorchestra is addressed. The mixing algorithm presented minimizes the difference between generation times of the media packets that are being mixed together in the absence of globally synchronized clocks, but in the presence of jitter in communication delays on packet switched networks. In order to support applications, such as teleorchestra, that involve a large number of participants, hierarchical mixing architectures are proposed, and it is shown that they are an order of magnitude more scalable than purely centralized or distributed architectures. Furthermore, mechanisms for minimizing the delays incurred by mixing in various communication architectures are presented. The mixing algorithms are implemented on a network of workstations connected by Ethernets, and the performance of various mixing architectures is experimentally evaluated. The results reveal the maximum number of participants that can be supported in a conference.>

The design of techniques and protocols for media mixing and communication architectures that optimize the performance of media mixing are discussed. The performance of monostage and multistage techniques for mixing are compared. Centralized and distributed architectures are analyzed for their suitability to both monostage and multistage media mixing, and hierarchical architectures that significantly reduce bandwidth consumption are presented. Algorithms for designing hierarchies that optimize real-time end-to-end delays and a packet train protocol in which the routing nodes can be used as mixers are proposed.>