J. Tárraga Giménez
Ph.D. dissertation by Joaquín Tárraga Giménez, under the supervision of Prof. Roger D. Hersch, Thesis No 2415 presented at the Computer Science Department, Ecole Polytechnique Fédérale de Lausanne, July 2001
Parallel servers for I/O and compute intensive continuous media applications are difficult to develop. A server application comprises many threads located in different address spaces as well as files striped over multiple disks located on different computers. This research presents new methods and tools to build parallel I/O- and compute-intensive continuous media server applications based on commodity components (PCs, SCSI disks, Fast Ethernet, Windows NT/2000). This dissertation describes a customizable parallel stream server based on a computer-aided parallelization tool (CAP) and on a library of striped file components enabling the combination of pipelined parallel disk access and processing operations. The parallel stream server offers a library of continuous media services (e.g. admission control, resource reservation, disk scheduling) to support parallel streaming. CAP allows to combine these continuous media services with application-specific stream operations in order to create efficient pipelined parallel continous media servers.
As an example of such a parallel continuous media server, this thesis presents the 4D beating heart slice server application, which supports the visualization at a specified rate of freely oriented slices from a 4D beating heart volume (one 3D volume per time sample). This server application requires both a high I/O throughput for accessing from disks the set of 4D sub-volumes intersecting the desired slices and a large amount of processing power to extract these slices and to resample them into the display grid. Parallel processing on several PCs and parallel access to many independent disks offers the potential of scalable processing power and scalable disk access bandwith. The presented 4D beating heart application suggests that powerful continuous media server applications can be built on top of a cluster of PCs connected to SCSI disks.
In order to extent the storage capacities of multi-PC multi-disk servers, this thesis also develops paradigms allowing to integrate optical jukeboxes into cluster-based server architectures. A scalable server based on optical disk jukeboxes offers highly accessible terabyte storage capacities for digital libraries, image and multimedia repositories with significant cost advantages. Server scalability can be achieved by increasing the number of optical disks, drive units, robotic devices and server PCs. The master server PC incorporates a shadow directory tree containing the directory trees of all present optical disk files. In order to offer higher throughputs, the scalable terabyte server allows to access large files striped over multiple optical disks loaded into different optical disk drives. Using magnetic disks as a cache for optical disk files allows to significantly increase the number of files that can be served at the same time.
Finally, this thesis presents a terabyte server comprising one master PC and two slave PCs, each slave PC connected to one NSM Satellite jukebox, each one comprising four optical disk reading units able to operate simultaneously. By running the 4D beating heart slice server application as well as Visible Human slice server application, the terabyte server offers advanced services in addition to its highly accessible terabyte storage capacities. The terabyte server has also been interfaced to a Web server and offers its multiple services to Internet clients. Part of this work has been published at the ACM Multimedia 1999 conference.
Download the full paper: PDF 5.4 MB