Portland State University Portland State University
PDXScholar PDXScholar
Computer Science Faculty Publications and
Presentations
Computer Science
11-1995
Quality of Service Speci=cation for Multimedia Quality of Service Speci=cation for Multimedia
Presentations Presentations
Richard Staehli
Oregon Graduate Institute of Science & Technology
Jonathan Walpole
Oregon Graduate Institute of Science & Technology
David Maier
Oregon Graduate Institute of Science & Technology
Follow this and additional works at: https://pdxscholar.library.pdx.edu/compsci_fac
Part of the Computer Engineering Commons, and the Databases and Information Systems Commons
Let us know how access to this document bene=ts you.
Citation Details Citation Details
Staehli, Richard; Walpole, Jonathan; and Maier, David, "Quality of Service Speci=cation for Multimedia
Presentations" (1995).
Computer Science Faculty Publications and Presentations
. 68.
https://pdxscholar.library.pdx.edu/compsci_fac/68
This Post-Print is brought to you for free and open access. It has been accepted for inclusion in Computer Science
Faculty Publications and Presentations by an authorized administrator of PDXScholar. Please contact us if we can
make this document more accessible: pdxscholar@pdx.edu.
Quality
of
Service
Specification
for
Multimedia
Presentations*
Richard Staehli, Jonathan Walpole and David Maier
{staehli, walpole, maier}@cse.ogi.edu
Department of Computer Science & Engineering
Oregon Graduate Institute of Science & Technology
20000
N.W. Walker Rd., PO Box 91000
Portland,
OR
97291-1000
ABSTRACT
The bandwidth limitations of multimedia systems force tradeoffs between presenta-
tion
data
fidelity and real-time performance. For example, digital video
is
commonly
encoded with lossy compression to reduce bandwidth
and
frames may be skipped during
playback to maintain synchronization. These tradeoffs depend on
device performance
and physical
data
representations
that
are hidden
by
a database system.
If
a multimedia
database
is
to support digital video and other continuous media
data
types,
we
argue
that
the database should provide a Quality of Service (QOS) interface
to
allow application
control of presentation timing and information loss tradeoffs.
This paper proposes a
data
model for continuous media
that
preserves device and
physical
data
independence.
We
show how
to
define formal QOS constraints from a
specification of ideal presentation outputs.,
Our definition enables meaningful requests
for
end-to-end service guarantees while leaving the database system
free
to optimize
resource management.
We
propose one set of QOS parameters
that
constitute a complete
model for presentation error and show
how
this error model extends the opportunities
for resource optimization.
Keywords:
Data
Model, Synchronization, Resource Reservations.
1
Introduction
Multimedia database systems are being extended to support presentations of continuous me-
dia
[8],
such as video and audio, as
well
as synthetic compositions such as slide shows and computer-
generated music.
We
call these presentations time-based because they communicate
part
of their
information content through presentation timing. While applications with text and numeric
data
types expect correct results from database queries, the real-time constraints of time-based presen-
tations commonly make
it
impossible to return complete and correct results. Some information loss
is also inevitable in any conversion
of
continuous media between analog and digital representations.
Consider the reproduction of NTSC video in a digital multimedia system.
The
video stream is
typically captured
at
640x480 24-bit samples/frame and
30
frames/second,
but
it is rarely stored or
played back
at
this bandwidth. Instead, lossy compression algorithms such
as
the MPEG encoding
[7]
are used
to
reduce the bandwidth requirements in exchange for some loss in quality. In addition,
if the display window
is
smaller
than
640x480 then the presentation will lose even more of the
This
research
is
supported
by
NSF
Grants
IRI-9223188
and
IRI-9111008,
and
by funds from Tektronix, Inc.
and
the
Oregon Advanced
Computing
Institute.
1
encoded
data
resolution. Contention for shared resources between applications also contributes
to
bandwidth restrictions and timing errors. Real-time MPEG players commonly drop late frames
rather
than
delay
the
remainder of the presentation.
Since
the
usefulness
of
time-based presentations depends on the accuracy of
both
timing and
data,
computing the result of a query in a multimedia database is a question
of
quality rather
than
correctness. Where database design has traditionally been concerned with the delivery of correct
results with acceptable delay, multimedia systems present a new challenge:
to
deliver results with
acceptable quality in real-time.
But
how accurate must a presentation be
to
be acceptable, and
how can
we
guarantee
that
a presentation achieves
that
accuracy? This paper helps to answer the
first question by giving a formal definition
of
presentation quality
that
measures
both
accuracy of
timing and the accuracy
of
output
values. This definition
of
presentation quality
is
then used
to
specify presentation-level QOS requirements.
The
second question can be answered by a variety
of
the techniques found in existing systems
[15].
However,
we
argue
that
current systems take an ad
hoc approach
to
presentation quality.
Without
a specification of presentation QOS requirements,
multimedia systems have no way of saying whether a presentation is acceptable or not.
As
a result,
these systems tend either
to
be overly conservative, wasting resources in an
attempt
to
guarantee
maximal quality; or overly liberal, accommodating resource shortages by unconstrained degradation
of
quality.
The
most common multimedia presentation tools use a best-effort approach, which aggressively
consumes resources
to
present all
data
as promptly as possible. When resources are overloaded,
a best-effort presentation will lose information. Many researchers have demonstrated best-effort
systems
that
maintain approximate synchronization despite variable latencies and resource avail-
ability
[6,
13,
2].
These systems show
that
a presentation can be acceptable even when quality
degradation is noticeable,
but
we
observe two problems with a best-effort approach. First, if per-
fect presentation is not necessary, why should a multimedia system expend extra "effort" for
the
best quality? Second, how much quality degradation can be allowed when many real-time presenta-
tions compete for scarce resources?
If
any application
is
to
be guaranteed acceptable service, some
information is needed
about
presentation QOS requirements.
Performance guarantees are an essential feature of real-time systems, including time-based pre-
sentations. While best-effort approaches offer only weak guarantees for synchronization, strong QOS
guarantees for continuous media presentations have been demonstrated through reservation
of
pro-
cessor, memory, network and storage system resources
[11,
19,
1,9,
12,3].
The
resource reservations
are derived from low-level
QOS parameters such as throughput, delay and
jitter
requirements for
stream processing. We call this a
guaranteed-best approach when reservations are based on require-
ments for a best-quality presentation.
The
primary problem with this approach
is
that
it
is
often too
expensive. For example, an instructional video with slow and deliberate motions may be digitized
and stored
at
30
frames/second,
but
playback at
15
frames/second is adequate for the purpose
of
instruction.
The
requirements for presentation quality depend not on the
data
type or view,
but
on
the
purpose of a presentation.
Others have recognized
that
best-quality presentations are often too expensive and unnecessary.
The
Capacity-Based-Session-Reservation-Protocol (CBSRP)
[17]
supports reservation of processor
bandwidth from the specification of a range
of
acceptable spatial and temporal resolutions for video
playback requests.
The
resolution parameters are intended only to provide a few classes
of
service
based on resource requirements and
not
to
completely capture presentation quality requirements.
Hutchinson et al.
[5]
suggest a framework of categories for high-level QOS specifications
that
include
reliability, timeliness, volume, criticality, quality
of
perception and even cost. They provide only a
partial list
of
QOS parameters
to
show
that
current QOS support in OSI and
CCITT
standards
is
severely limited.
The
error model
we
describe in this paper extends their approach
to
provide
a complete set
of
parameters
to
constrain presentation quality.
We
define presentation quality to
include only factors
that
affect perception of
the
information content
of
a time-based presentation.
Database technology offers many benefits for multimedia applications, such as high-level query
2
languages, concurrency, and device and physical
data
independence.
But
current database systems
do not adequately support time-based presentations. Relational
data
manipulation languages have
demonstrated the value
of
letting the application specify what is wanted, and letting the database
plan
how to retrieve it. To support time-based presentations, a
data
manipulation language for a
multimedia database should also allow the application to specify
when, where, and how precisely
the
data
should be delivered
[10].
These constraints on delivery are an example
of
a QOS-based
interface.
None of the proposed
data
models for time-based multimedia
that
we
are aware of support
queries for imprecise results. For example, Gibbs describes a
data
model
that
captures the structure
and synchronization relationships of complex time-based multimedia presentations
[4].
This model
includes media descriptors
that
attach a quality factor, such as "VHS quality" or "CD quality",
to each media object, but these labels describe the quality of the representation rather
than
the
presentation. Without the notion
of
presentation quality in the
data
model, one would presume
that
all information would be preserved in the result
of
a query. In practice, information loss in a
time-based presentation is inevitable and unconstrained by current
data
models.
This paper defines a methodology for presentation
QOS specification.
The
definitions are
intended to be general enough to apply to presentations in any multimedia system. In particular,
our methodology supports the following goals:
•
Model
user
perception
of
quality.
Just
as modern compression algorithms exploit knowl-
edge of human perception
[18],
a multimedia system can better optimize playback resources if
it
knows which optimizations have the least affect on perceived quality.
•
Formal
semantics.
We
would like to be able to prove
that
multimedia system can satisfy a
QOS specification.
•
Complex
data
model.
QOS specifications can be defined
for
a large class of complex mul-
timedia presentations.
The next section defines our terminology in terms of an architectural model for multimedia
presentations. Sections 3 and 4 describe a
data
model for the specification of content and view
respectively
for
a presentation.
We
then define quality in Section 5 as a function of a presentation's
fidelity
to
the content and view specification, in the context
of
an error model.
We
define one
possible error model, and suggest in
Section 6 how a formal QOS specification can be used to
optimize resource usage in a presentation.
Section 7 gives our conclusions.
2
Architectural
Model
In our architectural model, shown in Figure
1,
multimedia
data
come from live sources
or
from storage. Digital audio and video
data
have default content specifications associated with them
that
specify the sample size and
rate
for
normal playback. A time-based media editor may be used
to create complex presentations from simple content. A
player
is
used to browse and play-back
content specified by the editor. A user may control a player's
view parameters, such as window
size and playback rate, as
well
as quality parameters such as spatial and temporal resolution. The
combination of content, view, and quality specifications constitute a
QOS specification. When a
user chooses to begin a presentation, the player needs to verify
that
a presentation plan consisting
of real-time tasks will satisfy the
QOS
specification. A presentation plan
is
feasible if guarantees
can be obtained from a
Resource Manager for the real-time presentation tasks
that
transport and
transform the multimedia
data
from storage or other
data
sources to the system outputs.
This architecture
is
similar
to
other research systems
that
provide QOS guarantees based on
an admission test
[13].
However, our definition of QOS
is
novel in
that
we
make strong distinctions
between content, view
1 and quality specifications. A content specification defines a set
of
logical
image and audio
output
values as a function of logical time. A view specification maps content
onto a set of physical display regions and audio
output
devices over a real-time interval. Quality is
3
Figure
1:
An architecture for editing and viewing multimedia presentations.
video
cam 1 100-105
I cam2 50-53
audio
micl
10-25
o
5
cam1 108-115
8
15
time
...
Figure
2:
Timeline view
of
content specification for a presentation of bicycling video with audio.
a measure of how well an actual presentation matches
the
ideal presentation of content on a view
and a
quality specification defines a minimum acceptable quality measure.
We
will refer to quality
when
we
mean the measure, and QOS when
we
mean the combination
of
content, view, and quality
specifications.
By allowing independent control of content, view and quality, a multimedia system can offer
a wider range of services
that
take advantage of the flexibility
of
computer platforms. To illustrate
these services, consider the presentation of video and audio as described in Figure
2.
The
first
video clip refers
to
5 seconds of a digital video file.
The
video file
is
named
caml
because it was
captured with
the
first of two cameras recording the same bicycle racing event. The digital video
for
caml
has a resolution
of
320x240 pixels. A second video file named cam2 shows another view
of
the bicycling event and has a higher resolution of 640x480 pixels.
The
video presentation cuts from
caml
to cam!! for 3 seconds, and then back to
caml
for the last 7 seconds.
The
audio clip
file
mid
contains a digital audio sound-track recorded
at
the
same time as
the
video clips. After selecting
this content for presentation, a user should be able to choose view parameters and quality levels
independently. For example,
if
the
user chooses a view with a 640x480 pixel display window,
but
a quality specification
that
requires only 320x240 pixels of resolution, then the player may be able
to avoid generating the full resolution images from cam2.
The
quality specification allows
the
user
to indirectly control resource usage independent of the content and view selections.
The
player can
optimize resource usage
so
long as the presentation exceeds the minimum quality specification. Users
might also like to specify an upper bound on cost for resource usage,
but
since cost
is
independent
of information loss, constraints on cost are beyond the scope of this paper.
4
