Distributed Multimedia Systems

COMPONENTS OF A DISTRIBUTED MULTIMEDIA SYSTEM

i. A multi-user system designed to support multimedia applications for a large number of users consists of a
number of system components.
ii. Each system component serves a dedicated function and can be optimized for that function.
iii. A typical multimedia application environment consists of the following components:
Application software
Container object store
Image and still video store
Audio and video component store
Object directory service agent
Component service agent
User interface service agent
Networks (LAN and WAN)

Application software

i. The application software is the multimedia application that creates, edits, or renders (displays) multimedia
objects. The application functionality determines how multimedia objects are manipulated and the extent to
which the user can control the rendering of the multimedia objects.
ii. The application software performs a number of tasks related to a specific business process. The basic tasks
combined to form an application include the following:
Object selection: The user selects a database record or a hypermedia document from a file system,
database management system or document server.
Object retrieval: The application retrieves the base object which can be a customer record or a memo
depending on the nature of the application. Within the display of the base object, there may be some
buttons that allow the user to display or playback associated multimedia objects.
Object component display: Some document components are displayed automatically when the user
moves the pointer (mouse pointer or cursor) to the field or button associated with the multimedia
object. For example, clicking on a button for an image automatically brings up a window displaying
the selected image in it.
User initiated display: Some document components require user action before playback/display. For
example, an embedded video object requires the user to click on the button for the video object to
bring up a screen that simulates VCR controls. When the user pushes the play button, the video starts
playing.
Object display management and editing: Component selection may invoke a component-control
subapplication which allows a user to control playback or edit the component object. The example of
playing a video object is very applicable here. When the user pushes the button for the video object,
the subapplication for the display and playback of video object takes control and displays its own
screen simulating VCR controls. The subapplication may allow cutting and pasting multiple video
streams and sound tracks

A Container object store (or Document Store)

i. A container object store is used to store container objects in a network object server.
1

ii. The container may be a hypermedia document or a database record (or object).
iii. Typical repositories for container objects include relational databases, object oriented databases, and
document databases (such as Lotus Notes). Flat files and other database types can also serve this function.
iv. A document store is essential for applications that require storage of large volumes of documents. For
example, applications such as electronic mail, information repositories, & hypertext require storage of large
volumes of documents in document databases.
v. The following describes some characteristics of document stores:
Primary document storage: A file system or database that contains primary document objects
(container objects). Other attached or embedded documents and multimedia objects may be stored in
the document server along with the container object.
Linked object storage: Embedded components, such as text and formatting information,and linked
components, such as pointers to image, audio, and video components contained in a document, may be
stored on separate servers.
Linked object management: Link information contains the name of the component, service class or
type, general attributes such as size, duration of play for isochronous objects, and hardware and
software requirements for rendering.
vi. The increasing demand for hypermedia documents and linking documents with traditional data processing
database objects is giving rise to the need for designing systems that provide the infrastructure for locating
documents using pointers stored in objects in data processing databases and vice versa.

Image and still video store

i. An image/still video store is a mass storage component for images and still video. Document-images as

well as images for various other applications such as medical x-rays are stored for long durations and are
not editable.
ii. An image and still video store is a database system optimized for storage of images. Most systems employ
optical disk libraries called jukeboxes consisting of multiple optical disk platters that are played back by
automatically loading the appropriate platter in the drive under device driver control.
iii. The characteristics of image and still video stores are as follows:
Compressed information: The images are stored in compressed form. Typical compression factors
range from 20 to 50. The image must be decompressed before display.
Multi-image documents: Document images require another layer of linkage to identify images that
form the sequential pages of a document.
Related annotations: Both document images and other images such as those for medical x-rays may
have associated images that are superimposed for display, or may have image annotations.
Annotations may be stored as separate images or together with the main image in TIFF files.
Large volumes: Document image stores consist of large numbers of image files. Since each page is
stored as an image, an image repository for a large organization such as an insurance company or a
land registry can easily run into millions of scanned pages. These images need to be indexed and
stored. On a periodical basis, images may be migrated to slower on-line media or off-line storage
(archived), or purged.
Migration between high-volume media such as an optical disk library and high-speed media
such as magnetic cache storage: The usage patterns for images are understood quite well.
Maintaining on-line cache storage and migrating storage of image objects to slower near-line or offline media is critical for maintaining very large image databases.
Shared access: The image and still-video stores provide shared access to multiple users.
2

Audio and Full-Motion Video Store

Object Directory Service Agent

i. An audio/video component store is the storage resource used for storing audio and video objects.
ii. Audio and video objects are isochronous; i.e., they must be played back at a constant rate.
iii. A 30-second compressed sound clip is large. A 30-second full-motion video clip is even larger due to the
screen frame information. In both cases, lossy compression algorithms are used to compress data by
significant factors.
iv. The following lists some characteristics of audio and full-motion video object stores:
Large-capacity file system: A compressed video object can be as large as six to ten Mbytes for one
minute of video playback. Storing a 3D-minute video presentation or a speech can easily take up as
much as 200 Mbytes of storage.
Temporary or permanent storage: Video objects may be stored temporarily on client workstations,
servers providing disk caches, and multiple audio or video object servers. They may be purged or
migrated to more permanent near-line or off-line storage:
Migration to high-volume/lower-cost media: The video medium is very pervasive, & it will grow
into widespread use. Migration and management of on-line storage, near-line optical storage, and offline optical tape storage are of much greater importance and more complex than for images.
Playback isochronicity: Playing back a video object requires consistent speed without breaks. The
storage repository must be able to retrieve objects in a constant stream mode.
Multiple shared access: Objects being played back in a stream mode must be accessible by other
users. Different users may at any given time be accessing different sections of the object, that is, the
object must be able to play multiple streams that are not synchronized.

i. An object directory service agent is responsible for assigning identification for all multimedia object types
managed by that agent.
ii. The identification must be unique for the network and over time.
iii. The object directory agent service is then used by the component service agents for creating objects as well
as locating existing multimedia objects for linking with documents and database records. The directory
service is also used for retrieval and playback.
iv. It provides a directory of all multimedia objects on the server tracked by that element of the directory
service agent. The various elements of each class of object directory service agents must synchronize their
lists on a periodic basis.
v. The following describes the services provided by a directory service agent:
Directory service: The directory service lists all multimedia objects by class and server location.
Object assignment: A directory service agent also assigns unique identification to each multimedia
object. The identification must be unique throughout the network & must remain unique over time.
Object status management: The directory service must track the current usage status of each object
to ensure that the object is not archived or purged while it is being played back by a user workstation.
Directory service domains: The directory. service should be modular to allow setting up domains
constructed around groups of servers that form the core operating environment for a group of users.
The domain may map a complete operating facility of the corporation, a division, or even a
department.
Directory service server elements: Each multimedia object server must have an associated directory
service element that may reside on either the server or some other shared resource.
3

 Network access: The directory service agent must be accessible from any workstation on the network.
Direct access may be controlled at the domain level. Access outside the domain may be managed by
the directory service agent on behalf of a user workstation.
The directory service agent is a very important component of a distributed multimedia
system. It allows tracking each multimedia object, its replicated copies on the network, its current
use status, and its migration patterns.

Component Service Agent

User Interface Service Agent

i. A component service agent is responsible for locating each embedded or linked component object of a
multimedia container, and managing proper sequencing for rendering of the multimedia objects. For
example, a video object must be transmitted to the workstation at a fixed rate so it can be decompressed
and displayed at a constant rate. Associated soundtracks must start precisely at their cues (time markers).
ii. The component service agent is responsible for orchestrating the retrieval functions.
iii. Each multimedia component provides a service to the multimedia user workstation. This service may
consist of retrieving objects, managing playback of objects, storing objects, looking up objects to determine
on which server they may be, and so on.
iv. The following lists the characteristics and types of services provided by each multimedia component:
Object creation service: Component service agents obtain an identification for creating a new object
from the directory service agents and provide the user interface service agent access for storing the
new object captured/ created/ edited at the user workstation.
Playback service: Component service agents provide a set of standard services, such as play, seek,
search, copy, delete, and so on, for isochronous components.
Component object service agent: This is the code that provides these services for a specific object
type, such as a video component.
Service agents on servers: A component service agent cores ides wherever objects of its type are
stored, typically on all component servers, and temporarily on workstations when a temporary object
resides there. Multiple component agents may be coresident on a server if the server stores multiple
component objects.
Multifaceted services: Component objects may exist in several forms, such as compressed or
uncompressed. A component service agent can operate on an object in each of its forms, as well as
translate objects between forms. Translation may be more efficient on platforms that provide hardware
assistance.
i. [ A user interface services agent is responsible for managing the display windows on a user workstation,
interacting with the user, sizing the display windows, and scaling the decompressed object to the selected
window size.]
ii. The user interface service agent resides on each user workstation and provides direct services to the
application software for the management of the multimedia object display windows, creation and storage of
multimedia objects, and scaling and frame shedding for rendering of multimedia objects.
iii. The following lists the services provided by the user interface service agent:
Window management: Creates a new window for a multimedia object when invoked & registers it;
handles all messages for that window.
Object creation and capture: Requests component service agent to set up a new object, obtains the
identification for client application, and captures and stores new object.
Object display and playback: Sets up object for decompression; scales and adjusts frame speed for
display or playback of object.
4

 Services on workstations: The code for a user service agent usually resides on a workstation and
provides services to display or playback audio, video, image, or other multimedia components.
Using display software: The user interface service agent may be a thin layer on the Windows core
services or an X server. In any case, it must interact closely with the normal display manager of the
workstation.
iv. The user service interface agent should provide a well-defined API for all services required by the client
application. It may act just as a channel for some services required by the client application by transferring
requests to other service agents.

Networks (LAN and WAN)

It refers to the corporate-wide (or enterprise-wide) network consisting of all LAN and WAN interfaces
required for supporting a particular application for a specific group of users.

DISTRIBUTED CLIENT-SERVER OPERATION

i. The service agents in the multimedia application environment are combined to form a distributed clientserver system for multimedia applications.
ii. While the client-server architecture has been used for some time for relational databases such as Sybase
and Oracle, multimedia applications require functionality beyond the traditional client-server architecture.
For example, a directory service agent is not typically associated with traditional client-server architecture.
iii. The clients in this case were custom-designed for the server. Again it was assumed that the client-server
link was firmly established over the network, & that there was only one copy of the object on the specified
server.
iv. Figure below describes the client and server custom views in a large distributed database.

v. There are many advantages of the several custom views :

they provide the decoupling between the physical data and the user.
The physical organization of the data can be changed without affecting the conceptual schema by
changing the distributed data dictionary and the distributed repository.
Logical independence is achieved, and the conceptual schema can be changed without affecting the
external views.
vi. In multimedia databases, we have a combination of real-world data objects as well as projections in
images, sound, and video.
5

vii. The database must assign some form of identification and an understanding of the data. Furthermore, on
retrieval these objects may need special processing before being rendered on user screens.
The different issues in distributed client-server operations are:
Clients in Distributed Workgroup Computing
Servers in Distributed Workgroup Computing
Middleware in Distributed Workgroup Computing
Database Operations

Clients in Distributed Workgroup Computing

Servers in Distributed Workgroup Computing

Database Operations

i. Clients in distributed workgroup computing are the end users with workstations running multimedia
applications.
ii. These client systems interact with the data servers in any of the following ways:
Request specific textual data
Request specific multimedia objects embedded or linked in retrieved container objects
Require activation of a rendering server application to display /playback multimedia object
Create and store multimedia objects on servers
Request directory information on locations of objects on servers
iii. In true distributed operation, the clients have no specific knowledge of where the data servers are and how
the data is organized.
iv. In true distributed object computing, the user is primarily concerned with the data object and its
manipulation relative to other data objects, rather than storage locations or applications that provide the
manipulation functionality.
i. Servers, in addition to the basic function of storing data objects, provide a number of other functions,
including those listed as follows:
Provide storage for a variety of object classes
Transfer objects on demand to clients
Provide hierarchical storage for moving unused objects to near-line (optical disk libraries) or off-line
(optical tape libraries) media
System administration functions for backing up stored data
Direct high-speed LAN and WAN server-to-server transport for copying multimedia objects
ii. In addition to these functions, advanced object server systems provide functions such as ensuring that
sufficient copies of data objects are available to meet user throughput requirements, replicated copies of
data objects remain synchronized, and the user perceives the distributed storage system as a single storage
entity.
Most database systems are used to perform a basic set of operations. These include the following:

Search

In conventional non-multimedia databases, search operation is to find an object (record) in response to a

query.
However, in the case of distributed multimedia objects, searching may involve additional transactions to
locate a copy of the required object and to obtain a copy from a remote server onto a local server for
further client operations.
6

 It is further possible that the query may require special programs to scan the multimedia object to
recognize components in it that may be required or set up as the starting point. For example, in JPEG
and MPEG compressed audio and video files, index markers or DCT scene-detection processing can be
used to jump to successive scenes and start retrieval at a specific scene.

Browse

In information & document databases, browse function is much more useful.

It may be required not only to retrieve attribute information about the objects but also renders frame of
the object contents.

Retrieve

Retrieve functions are different for images, audio, and video from symbolic text-only databases because
all three multimedia objects require the retrieved data to be processed by specialized decompression
engines before being rendered.
An important issue is whether the decompression is performed at the server or the client, or in a
specialized decompression server .

Create & Store

Create and store functions in distributed relational databases are concerned primarily with finding the
tables in which the data has to be stored and updating distributed storage indexing information.
In the case of multimedia objects, the objects themselves are not stored alongside the field data. Rather,
the objects are stored in a separate server and require an object directory to provide indexing information
for retrieval.

Update

< No useful information in book Multimedia Systems Design by Prabhat K. Andleigh & Kiran Thakrar,
Pearson Education >

Middleware In Distributed Workgroup Computing

i. The primary role of middleware is to link back-end database servers to front-end clients in a highly flexible
and loosely connected network model.
ii. A loosely connected network model implies that servers may go off-line and be unavailable without
bringing the network down or significantly impacting overall operation.
iii. Also, clients may go off-line temporarily and continue local operations. When they connect later, they can
operate as a part of the network & resynchronize their databases.
iv. Middleware dynamically redirect client requests to appropriate servers that are on-line, thereby also
providing a potential load-balancing function under demanding conditions.
v. Middleware performs a number of functions in this environment:
Provide the user with a local index, an object directory, for objects with which a client is concerned
Provide automatic object directory services for locating available copies of objects
Provide protocol and data format conversions between the client requests and the stored formats in the
server
Provide unique identification throughout the enterprise wide network for every object through time
vi. The database architecture changes significantly when middleware is introduced in the system.
vii. The middleware is capable of accessing multiple databases and combining information for presentation to
the user. For example, middleware can perform some or all combinations of the following functions:
Access a document database to locate a pointer to the required multimedia object
7

 Locate an object using a distributed object directory database

Access an object database to retrieve an object
Retrieve object preprocessing information from an object description database
Combine all of this information and preprocess the object before passing it on to a client
viii. These actions of middleware content-based processing. The range and nature of such content-based
processing can be changed without affecting either the servers or the clients.
ix. Content-based processing allows the middleware to address temporal characteristics of certain multimedia
objects such as audio and video. It also allows a variety of editing and updating functions on stored
multimedia objects.

Role of Middleware in Distributed Multimedia Operation

MULTIMEDIA OBJECT SERVERS

i. Multimedia systems consist of a number of information objects, including text, binary files, images, voice,
and full-motion video shared by a number of users or, in the case of electronic mail (e-mail), are routed
from one user to another.
ii. To achieve this functionality, the information objects must be stored on network resources accessible to all
users who need to access them.
iii. The resources where information objects are stored so that they remain sharable across the network are
called servers.
iv. These information object servers may reside on file servers dedicated to a single class of objects or share
the file server with other object servers.
The different issues concerned with multimedia object server are :
Types of Multimedia Servers
Mass Storage for Multimedia Servers
Network Topologies for Multimedia Object Servers

Types of Multimedia Servers

i. Separation of-objects & storage is an effective means of ensuring performance of the underlying
application that uses the numerical and textual data first.
8

ii. The requirement that all the data whether alphanumeric, images, audio, or video, be an integral part of the
established database management framework at the same time; is complex because of the sheer size these
data objects.
iii. An image, audio, or video objects are so large in size that they are too unwieldy (-difficult to control) for
conventional databases and, when combined with alphanumeric data objects, may reduce the overall
database performance.
iv. Thus, they need separate servers to ensure performance by customizing each server for the type of objects
stored in it. Separate servers also makes server management easier.
v. In an ideal case, each object type would have its own dedicated server optimized for that type of object.
Hence, the number of servers depends directly on the types of data objects supported by the multimedia
system.
vi. At the very least, a network would consist of some combination of the following different types of servers:
Data-processing servers supporting RDBMSs and ODBMSs
Document database servers
Document imaging and still-video servers
Audio and voice mail servers
Full-motion video servers

Data-processing servers (RDBMSs or ODBMSs)

i. These are traditional database servers that contain alphanumeric data.

ii. In a relational database, data fields are stored in columns in a table.
iii. In an object-oriented database, these fields become attributes of the object.
iv. In either case, indexing some fields or attributes is essential for fast access to data.
v. The databases are designed for rapid searches of objects using one of the indexed fields or attributes.
vi. The database serves the purpose of organizing the data and providing rapid indexed access to it.
vii. The database management system can interpret the contents of any column or attribute for performing a
search.

Document database servers

i. These used for electronic mail databases and for document-based information repositories.
ii. They are similar to conventional database storage as they are also predominantly alphanumeric and may
contain some indexed alphanumeric fields.
iii. They differ from conventional databases as they contain special text fields that may be indexed within
themselves using a hypertext engine.
iv. In addition, text fields in document databases that support hypermedia documents may have embedded or
linked binary files, images, audio, & video objects.

Document imaging and still video servers

i. They store and manage image and still-video objects.
ii. These objects may be in the form of basic operating-system-level files, or server files indexed in
some manner for rapid location of the required image.
iii. In an object database, they may be indexed persistent objects.
iv.A file or an object may contain a one page image object (or file) or a complete document
consisting of multiple pages.
v. The server software may be set up with special caching mechanisms to speed up access to images.
Audio and voice mail servers

i. They are used primarily for applications such as voice mail,voice annotations, and voice help messages.
ii. Audio objects are large even in compressed form.
iii. Audio servers may serve two different types of applications:
9

 Traditional telephone-based voice mail, and

Voice mail messages linked with the document based messaging system.

Full-motion Video servers

i. They are designed to manage very large objects. For example, a 10-second video object requires over a
megabyte of storage.
ii. Besides providing the usual indexing functions, video servers are made intelligent to support the
isochronous playback requirements for video objects by reserving network bandwidth.

Mass Storage for Multimedia Servers

i. Mass storage is an enabling technology that is fueling rapid growth during this multimedia information age.
ii. User demand for storage management of information objects is served by several storage technologies:
battery-powered RAM, nonvolatile flash memory, rotating magnetic disk drives, and rotating optical disk
drives.

Magnetic Disks

i. Rotating hard-disk-drive magnetic storage is the most broadly used technology for information
management applications.
ii. Rotating magnetic and optical storage continue to be the storage of choice for large-volume information
objects due to continual reduction in the price per megabyte of high-capacity storage, because of continual
increase in areal recording density (megabit/square inch), without any sign of slowing.
iii. Over a ten-year period, the price per Mbyte for magnetic hard disk storage has dropped by a factor of ten.
iv. During the past decade, disk drive sizes have declined from 8-inch and 5.25-inch form factors to 3.5-inch
and 1.8-inch drives. These small drives make it possible to have high volume storage in notebook &
subnotebook computers.
v. Smaller 1.3 inch & sub-1-inch disks make it possible to build credit-card-sized drives that fit in the
Personal Computer Memory Card International Association (PCMCIA) memory-card form factor. Such
memory-card drives provide removable hard-disk-drive storage for subnotebook and handheld portable
computers, providing a higher degree of data transportability and control.
vi. Reductions in drive size have been accompanied by continual increases in ruggedness. Shock and vibration
capability for an operating 1.8-inch drive are on the order of 10 g (a measure of acceleration). Several new
sub-2-inch drives claim operating shock of 100 g.
vii. The evolution of magnetic disk drive technology is important from two perspectives:
The higher storage capacities, and
redundant drive arrays
These perspectives provide the option of using magnetic drives rather than optical drives for most on-line
storage of multimedia objects.
viii. Thus, magnetic drives can serve multiple roles as
networked multimedia object servers,
temporary caches for multimedia objects as they migrate from one kind of storage to another, and
local multimedia object servers on user workstations (both desktop and portable notebooks)

RAID

i. RAID promises a high degree of fault tolerance-a necessity for networking environments supporting
workgroup applications.
ii. They also reduce bottlenecks and provide higher bandwidths, key issues for full-motion video systems.
iii. In terms of redundancy, RAID provides a more cost-effective solution, although lower performance.
10

iv. RAID systems use multiple and potentially slower disks to achieve the same task as a single expensive
large capacity and high transfer rate disk. In RAID, high transfer rates are achieved by performing
operations in parallel on multiple disks.
v. These technologies are combined in proprietary higher-level RAID implementations to achieve specificfunctionality isochronous data transfers.
vi. Of these, the following six levels have been defined, and RAID Level 6 is being standardized:
Disk striping (Level 0)
Disk mirroring (Levell)
Bit interleaving of data (Level 2)
Byte interleaving (Level 3)
Sector interleaving (Level 4)
Block interleaving (Level 5)
vii. The type of application & how manages the data determines the type of RAID technology that is optimum.
The typical block sizes and required response times determine the type of RAID technology that provides
desirable efficiency and reliability.
viii. RAID technology can be implemented in hardware or software.
Hardware techniques use a dedicated processor on the controller board to sequence the disk writes or
reads, and to perform parity creation on disk writes and checks on disk reads.
Software implementations use the host CPU to perform the dame functions
A hardware solution is better, but it is more expensive.
ix. When using RAID, it is important to keep in perspective the effects of RAID on disk writes and reads for
the variety of applications that will be used on the system. RAID technology provides high performance for
disk reads for almost all types of applications, a key requirement for streaming video at a constant rate.

Write-Once Read-Many (WORM) Optical Disks

i. They provide very high volumes of storage for very low cost.
ii. A 12-inch WORM drive can store more than 6 Gbytes of information.
iii. At an average of 50 Kbytes per compressed document page, a disk platter can store over 120,000 document
images.
iv. A 5.25-inch platter can store as much as 1 Gbyte of information.
v. Some key characteristics are important to note here:
Optical drives tend to be slower than magnetic drives by a factor of three to four.
WORM drives can write once only; typically, 5 - 10% of disk capacity is left free to provide for
changes to existing information. They are useful for recording information that will not change very
much.
They are virtually indestructible in normal office use and have long shelf lives.
They can be used in optical disk libraries (jukeboxes). A jukebox may provide anywhere from 50 to
100 disk platters with two or more drives.
vi. These characteristics make optical disks ideal candidates for on-line & near-line document images, still
video objects & archived data, but they are not best suited for video objects. They can serve video objects
well only in the role of an archival media that can ultimately be removed from the jukebox for shelf
storage.

Rewritable Optical Disks

i. Magneto optical and other similar technologies have been used to produce rewritable optical drives.
ii. One disadvantage they have common with WORMs is slow data transfer rate. Compared to WORM disks,
rewritables have a considerable advantage due to rewritability.
11

iii. They can be used as primary or secondary (near-line) media for storage of large objects, which are then
archived on WORM disks.
iv. When used as primary media, they must be used in conjunction with a high-speed magnetic disk cache to
achieve acceptable video performance.
v. Secondary media refers to media used to store objects that have passed their three-month to six-month
initial phase and can be delegated to slower near-line media.
vi. These objects can be removed from this slower media when they are archived.

Optical Disk Libraries

i. Both WORM disks and rewritables can be used in optical disk libraries (jukeboxes) to achieve very high
volumes of near-line storage.
ii. A key disadvantage of optical disk libraries is the time it takes for a platter to be loaded into a drive and
spun to operating speed. Depending on the disk library, this delay can be as high as 10 to 20 seconds.
iii. Applications designed to use information objects located in optical disk libraries are designed to adjust for
this delay by prefetching predicted objects into magnetic disk cache.

Network Topologies for Multimedia Object Servers

i. The isochronous nature of full-motion video plays an important role in the selection of network topologies
that can be used for multimedia systems featuring full-motion video.
ii. A number of topology options can be viewed as potential solutions depending on the application, the size
of the organization, the topology of the distributed systems, and the locations of various data servers.
iii. The multimedia object servers are dedicated to one or more multimedia objects-graphics, document
images, audio, and full-motion video.
iv. Three different approaches to setting up multimedia server are as follows :
Centralized Multimedia Server
Dedicated Multimedia Server
Distributed Multimedia Server

Centralized Multimedia Server

i. A centralized multimedia object server performs as a central store for multimedia objects.
ii. All user requests for multimedia objects are forwarded by the applications to the centralized server and are
played back from this server.
iii. The centralized server may serve a particular site of the corporation or the entire enterprise.
iv. The advantages of this approach are for a small enterprise with fewer than 100 active user.
v. This topology provides access to all users on any LAN in the enterprise. The performance for each user is
dependent on the' number of LAN traversals & the switching latency for each lap.
vi. For a very large enterprise with a number of geographically distributed facilities, performance is a more
serious issue.
vii. The disadvantages of a centralized data server for a large enterprise are obvious. Users linked to the server
via the WAN are dependent on the speed provided by the WAN.
viii. A WAN supporting 56 Kbits/sec is not capable of playing an MPEG compressed video at 1.5 bits/sec
without serious degradation in quality.

Dedicated Multimedia Servers

i. If a video server is on a separate dedicated segment, there is no other contention within that segment for
LAN traffic.
ii. When a workstation dumps a large video, the other servers on the network are not affected. From a
performance perspective, this approach provides very high performance for all local operations.
iii. The isochronicity of audio & video objects is handled quite well in a dedicated mode.
12

iv. A major disadvantage of this approach is the level of duplication of objects. Every dedicated multimedia
object server has to have its copy of every multimedia object required by a workstation local to the LAN.
v. This requirement can become an object management nightmare and may place an unnecessary load on the
network services.

Distributed Multimedia Servers

i. This approach falls somewhere between the centralized & dedicated approach.
ii. The problem of duplicating multimedia object on each server is mitigated by managing multimedia object
servers in a more intelligent manner by distributing them in such a manner that they are placed in strategic
locations on different LANs and replicate on a programmed basis to provide balanced service to all users.
iii. While the distributed multimedia server approach addresses the key issues of performance and
expandability, this approach also presents challenges in maintaining current information for all users and
ensuring that all users have access to information they need.

MULTISERVER NETWORK TOPOLOGIES

i. Multiple object servers for different classes of information objects, distributed across the enterprise, must
operate in consort to provide a user with a complex multimedia object on demand.
ii. Multimedia application using complex objects consisting of consisting of some combination of text, image,
voice and video place a substantial strain on network topologies.
iii. The time dimension of full-motion video (isochronicity) requires extremely short latency times in network
switching.
iv. A number of different network topologies have been tried and are used. The primary topologies include the
following:
Traditional LANs (Ethernet or token ring)
Extended LANs (using network switching hubs, bridges, and routers)
High-speed LANs (ATM and FDDI II)
WANs (including LANs, dial-up links-including ISDN, Tl, and T3 lines etc.)

Traditional LANs

Extended LANs

i. Traditional Ethernet LANs operate at 10 Mbits/sec and token ring LANs at 16 Mbits/sec.
ii. A typical compressed video at 1.5 Mbits/sec is considered adequate for most office use. LANs at this speed
can support a number of simultaneous sessions in a mix of live video, audio, electronic mail, and so on.
iii. It is conceivable that a LAN such as this can support anywhere from 5 to 10 simultaneous users and can
allow as many as 20 to 30 connections on the LAN before performance becomes a problem
i. Use of hubs, bridges, and routers address some of the issues of higher network bandwidth by combining a
number of separate LAN segments into a continuously addressable LAN.
ii. In extended LANs, each segment operates at the normal LAN bandwidth.\
iii. Most medium-sized to large organizations have been using high-speed fiber optic networks as backbones to
connect departmental LANs.
iv. Object servers may reside on their own LAN or may be distributed across departmental LANs.
v. This is a design issue that must be resolved on the basis, how often other departments use objects from the
object server.
13

Switching Hub

i. Hub are used to switch one LAN segment (or strand) to another LAN segment with very low latency. For
example, a hub may support 64 Ethernet strands with the potential of 32 simultaneous connections.
ii. This approach can be used to address the isochronicity of the full motion video.
iii. These networks can support the requirements for full-motion video standards such as MPEG2. Each
individual LAN segment can support five concurrent DVI streams operating at 1.5 Mbits/sec.
iv. An important advantage of this approach is that the user workstations do not require additional LAN
hardware if they are already connected to a LAN.
v. The user workstations continue to operate on low-cost lower-speed LAN connections that also fully
support all other applications.
vi. The switching hub scheme has been used to enhance 10Base-T Ethernet circuits for supporting full-motion
video.
vii. Other schemes use shared memory buffer store-and-forward methods in addition to switching, thereby
combining switching functions with bridging & routing.
viii. Switching hubs can operate at speeds ranging from 100 Mbits/sec to over 1 Gbits/sec. At 1 Gbits/sec a
switching hub can accommodate a very large number of simultaneous data streams at 10 Mbits/sec (for
Ethernet connections) or at 1.5 Mbits/sec(full-motion-video transmission rate).
ix. These hubs can also act as gateways to carrier ATM networks and FDDI LANs.

Bridges and Routers

i. Bridges and routers differ from hubs.

ii. Hubs switch (and connect) one LAN segment to another LAN segment, bridges and routers transfer a
packet of data from one LAN segment to another LAN segment.

Switching and Routing Latency

i. Every internetworking device has some level of delay caused by the processing within the device to
determine the source and target network segments and then switch the packet to the target segment.
ii. Switching latency is defined as the time it takes a switching hub to interconnect one LAN segment to
another LAN segment.
iii. While switching latency is not very visible in application using just data transfers, switching latency
becomes very irritating to a user viewing a video or listening voice mail messages.
iv. Routing delay is defined as the delay experienced by a packet of data within the router. For ordinary data
translation, this shows up as a slow network. For multimedia applications, this delay shows up as voice that
sounds staccato and video that moves in fits and starts, and the picture has no synchronization with the
sound.
v. When a network transaction traverses several internetworking devices, the delays add up. This total delay
must be considered as the internetworking latency.
vi. The following lists the typical delays in the various internetworking components:
PC-based software routers: 20,000 microseconds
Hardware routers:
1,000 microseconds
Bridges:
Greater than 400 microseconds
Switches:
Greater than 40 microseconds

High-Speed LANs

i. High-speed LANs such as FDDI II and ATM can support a much larger number of users.
ii. For example, FDDI II-with speeds ranging 'from 100 Mbits/ sec to 300 Mbits/sec can support a couple of
hundred users.
iii. FDDI II is a single-media LAN and its full bandwidth supports all users.
14

iv. ATM is a switched LAN like a PABX, and each user can use the bandwidth for their individual connection.
v. The switching speed of the ATM hub determines the overall capacity of the ATM network. Depending on
the hardware, ATM networks can support individual connections as high as 155 Mbits/sec and hub speeds
of 622 Mbits/sec.
vi. FDDI II is not a good candidate for LANs directly connecting workstations, because :
Each user workstation requires a very-high-speed network interface
It does not support workstations operating at different speeds; separate LANs operating at different
speeds are required for that.
vii. ATM is actually a good candidate for two reasons: as a hub-and-spoke technology, it adapts very well to
the wiring closet paradigm; and it allows workstations to operate at speeds defined by the workstation.
viii. ATM appears to be more promising than FDDI II due to the ability to operate user workstations at lower
speeds while the switch itself operates at a much higher speed.
ix. This is an important consideration because it allows object servers to operate at high speeds to support
multiple requests, while workstations can operate at the class of service for which they are designed.
x. Object servers need to replicate objects and transfer objects at high speed." on demand. FDDI II appears to
be a very useful high-speed technology for connecting servers on an additional separate network and
providing the dedicated high bandwidth necessary for rapid transfer and replication of information objects.

WANs

i. WANs, by our definition, include LANs, dial-up ISD~, T1 (1.544 Mbits/sec) and T3 (45.3 Mbits/sec) lines,
and regular telephone dial-up lines.
ii. WANs may have a mix of networking and communications protocols.
iii. WANs have a variety of speeds at which various parts of it communicate.
iv. In a WAN, any communication across the WAN may have to traverse both a variety of protocols as well as
different speeds of the WAN.
v. From an application perspective, dealing with a wide range of protocols and speeds can be a management
disaster that might make the application very complex.
vi. An application can be kept simple by hiding the complexity of the WAN from the application protocol
layering.
The application interacts with a higher-level layer of the protocol and remains unaffected by changes
in the lower layers of the protocol.
The lower-level protocols are then free to address the changes in the transmission medium.

15