HERMES: Distribution and Caching for Multimedia Data |
This project aims at developing high-performance multimedia information-system technology. |
This project aims at developing high-performance multimedia information-system technology. Our focus in this sub-project is on distributed caching strategies for continuous media in application areas such as medical information systems, the virtual campus or news-on-demand. We assume a computing infrastructure consisting of a number of servers and a moderate-to-large number of clients connected by a tightly-coupled cluster of local-area networks. Cumulative client resources may exceed server resources (CPU's, and primary and secondary-memory capacity and bandwidth). Many data placement and scheduling algorithms for continuous media in single-disk(e.g. scan-EDF) or in multi-disk(e.g. staggered striping) systems are round-based. Round-based schemes assume that retrieval, transmission and display activities are all synchronised on a round-by-round basis. In the architectural context described above, there are a number of grounds for considering round-based schemes too restrictive: |
Capacity: |
A consequence of being round-based is that round-length must be tuned to the weakest network-connections and clients in the system. That is, the resources of faster connections and of larger clients cannot be fully exploited. |
I/O Bandwidth: |
Longer round lengths have several advantages for placement and I/O bandwidths. These include reduced overheads due to seek and rotational delays, and increased reward from effective intra-stream data placement strategies. |
Mixed Workload: |
Multimedia systems must accommodate more than continuous media only: a variety of classes of non-continuous media including images and text must also be accommodated, in addition to structured data and access structures. The assumption that a significant proportion of all resources be set aside for round-based scheduling and processing, therefore, is somewhat restrictive. |
For these reasons, we are developing scheduling and caching algorithms between servers and clients which aim to increase the capacity of multimedia servers and systems as a whole through effective use of client resources. The main client resources we consider are primary memory resources, secondary memory resources and secondary memory bandwidth. Particular strategies are: |
Prefetching: |
That is, the early scheduling and retrieval of continuous data, and client caching. Comparing MPEG-2 bandwidth requirements with those offered by modern disk systems, client caching on secondary storage is feasible. |
Multicasting: |
Related to prefetching, multicasting is the broadcast of data across the network to all clients which will (at some point) need the data. This also requires large (secondary) caches. |
Multi-sourcing: |
Given significant prefetching and caching between clients, the possibility to service new client requests from other client caches arises, thereby removing that load from servers. |
Implementation tasks within this project are being carried out within the prototype system. |
Project Partners: |
This project is part of the ESPRIT project (9141) with partners in Greece (MUSIC/TUC), Germany (GMD-IPSI) and Italy (CNR-IEI) and aims at developing high-performance multimedia information-system technology. |
Funding: |
Our part in this project is funded by the BBW (93.0135), and the project as a whole is funded under the ESPRIT initiative (9141). ETH works, in this project, in close collaboration with Prof. Gerhard Weikum (University of Saarbrücken). |
contacts: Prof. H.-J. Schek |