Hp COMPAQ PROLIANT 3000, COMPAQ PROLIANT 5500, COMPAQ PROLIANT 1600 Video Streaming Technology

WHITE PAPER

.

July 1998

Compaq Computer Corporation

ECG Emerging Markets and Advanced Technology

CONTENTS

VIDEO TECHNOLOGY.............. 3

The Human Eye.......................3

Analog Video ............................ 4

Analog Composite Video..............4

Analog Component Video ............4

Digital Video.............................5

Digital Video Formats ..................5

Network Delivery Challenges.......6

The Bandwidth Problem............6

Scaling........................................ 7

Compressing—Codecs................ 8

Video Codec Standards ............ 9

H.261.......................................... 9

H.263.......................................... 9

JPEG and MJPEG ....................... 9

MPEG.......................................10

VIDEO-STREAMING..............12

Isochronous Video...................13

Video Streaming System.........14

Network Considerations...........15

LANs/Intranets .......................... 17

Public Internet...........................18

Public Broadband Networks.......20

VIDEO SERVERS..................22

Application Software................22

Video Server Hardware............23

High Capacity Disk Storage.......24

High Sustainable Throughput.....24

High Performance Network........24

Multiple CPUs ........................... 24

Expandable System Memory .....24

High Availability.........................24

Rack-mount for Easy Access ..... 25

Attractive Cost per Stream......... 25

Example: Compaq ProLiant.....25

ACRONYMS......................... 27

RESOURCES .......................28

1ECG068/0798

.

Video Streaming Technology

.

If a picture is worth a thousand words, then a video is worth a thousand pictures. The

.

sights and sounds of video teach us, entertain us, and bring our fantasies to life. While

.

text, graphics, and animation provide for interesting content, people naturally gravitate

.

to the richer and more realistic experience of video. That is because video—with audio—

.

adds the ultimate level of realism to human communication that people have come to

.

expect from decades of watching moving pictures in the real-world media of TV and

.

movies.

.

As all such real-world media continues to migrate toward "everything digital", video too

.

is becoming digital. Video delivery has evolved from the analog videotape format of the

.

1980s to a digital format delivered via CD-ROM, DVD-ROM and computer networks. As

.

a series of digital numbers, digital video has the advantage of not degrading from

.

generation to generation, and, because it can reside on a computer disk, it is easy to

.

store, search, and retrieve. It can also be edited and easily integrated with other media

.

such as text, graphics, images, sound, music, as well as transmitted without any loss in

.

quality. And now it is possible to deliver digital video over computer networks including

.

corporate Intranets and public Internets directly to desktop computers.

.

What makes this network delivery possible is the emergence of new technology called

.

“video-streaming”. Video streaming takes advantage of new video and audio

.

compression algorithms as well as new real-time network protocols that have been

.

developed specifically for streaming multimedia. With video streaming, files can play as

.

they are downloaded to the client, thus eliminating the necessity to completely download

.

the file before playing, as has been the case in the past. This has the advantages of

.

playing sooner, not occupying as much disk space, minimizing copyright concerns, and

.

reducing the bandwidth requirements of the video.

.

This white paper discusses the salient characteristics of this new video-streaming

.

technology. How the properties of human vision shape the requirements of the

.

underlying video technology. How the high bit rate and high capacity storage needs of

.

video drives the demand for high video compression and high bandwidth networks. How

.

the real-time nature of video demands the utmost in I/O performance for high levels of

.

sustained throughput. How the characteristics of video streaming shapes the

.

requirements of video server hardware. And finally, how Compaq video streaming

.

servers meet these demanding requirements through high performance I/O architectures

.

and the adherence to industry standards.

.

Please direct comments regarding this communication to the ECG Emerging Markets and Advanced Technology Group at:

.

EMAT@compaq.com

.

WHITE PAPER (cont.)

.

NOTICE

.

THE INFORMATION IN THIS PUBLICATION IS SUBJECT TO CHANGE WITHOUT

.

ECG068/0798

.

NOTICE AND IS PROVIDED “AS IS” WITHOUT WARRANTY OF ANY KIND. THE

.

ENTIRE RISK ARISING OUT OF THE USE OF THIS INFORMATION REMAINS WITH

.

RECIPIENT. IN NO EVENT SHALL COMPAQ BE LIABLE FOR ANY DIRECT,

.

CONSEQUENTIAL, INCIDENTAL, SPECIAL, PUNITIVE OR OTHER DAMAGES

.

WHATSOEVER (INCLUDING WITHOUT LIMITATION, DAMAGES FOR LOSS OF

.

BUSINESS PROFITS, BUSINESS INTERRUPTION OR LOSS OF BUSINESS

.

INFORMATION), EVEN IF COMPAQ HAS BEEN ADVISED OF THE POSSIBILITY OF

.

SUCH DAMAGES.

.

The limited warranties for Compaq products are exclusively set forth in the documentation

.

accompanying such products. Nothing herein should be construed as constituting a further or

.

additional warranty.

.

This publication does not constitute an endorsement of the product or products that were tested.

.

The configuration or configurations tested or described may or may not be the only available

.

solution. This test is not a determination of product quality or correctness, nor does it ensure

.

compliance with any federal state or local requirements.

.

Compaq, Deskpro, Compaq Insight Manager, ProLiant, Netelligent, and SmartStart are registered

.

with the United States Patent and Trademark Office.

.

Microsoft, Windows, Windows NT, Windows NT Server 4.0, Terminal Server Edition, Microsoft

.

Office 97, Microsoft Excel, Microsoft Word, Microsoft Outlook 97, Microsoft Internet Explorer,

.

PivotTable, Microsoft Visual Basic, Microsoft SQL Server, Microsoft Exchange, and Internet

.

Information Server are trademarks and/or registered trademarks of Microsoft Corporation.

.

Pentium, Pentium II, and Pentium Pro are registered trademarks of Intel Corporation.

.

Other product names mentioned herein may be trademarks and/or registered trademarks of their

.

respective companies.

.

Video Streaming Technology

.

First Edition (July 1998)

.

ECG068/0798

.

2

WHITE PAPER (cont.)

.

Visible Light

Rays

.

Video Technology

.

ECG068/0798

.

The Human Eye

.

An understanding of video technology starts with an understanding of the properties of the human

.

eye. This is because the electronic eye of the video camera tries to mimic what the human eye

.

sees. Basically, the human eye detects, or “sees”, electromagnetic energy in the visible light

.

frequency spectrum ranging from a wavelength of about 400 nanometers (nm) to 700 nm. The

.

eye cannot detect electromagnetic radiation outside this spectrum.

.

8000nm .0001nm

.

3

TV

Waves

The human eye detects this light energy through the use of photoreceptors known as "rods" and "cones". There are approximately 120 million rods distributed across the spherical surface called the "retina" at the back of the eye. These rods are sensitive to light intensity. By contrast, there are only about 8 million cones that are sensitive to color on the surface of the retina. These cones tend to be centrally located at the back of the eye. The large number of light-sensing rods, compared to color-sensing cones, makes the eye much more sensitive to changes in brightness than to changes in color. This fact is taken advantage of in the video and image compression schemes discussed below which sample color at a lower rate than that of brightness. This is also why night vision, which relies on the low light intensity sensing capability of rods, is devoid of color, and why peripheral vision, which is not directed at the center of the retina is not as color sensitive.

Brightness changes

detected by 120M rods

For color sensing there are three types of cones capable of detecting visible light wavelengths. It has been determined that a minimum of three color components—e.g., Red, Green, and Blue— corresponding to the three types of cones, when properly filtered, can simulate the human sensation of color. Since color does not exist in nature—it is literally in the eye and brain of the beholder—these cones sense light in the visible spectrum and our brain processes the result to provide us with the sensation of color. This process is additive in that the brain can create colors—e.g., red + blue = purple—that don’t exist in the pure spectrum. These properties of human vision are used in video compression schemes, as well as in display systems to provide efficient methods for storing, transmitting, and displaying video data.

Another characteristic of human vision important to video technology is that of “image persistence”. This is where an image remains on the retina, even though the original object has been physically removed or replaced by another image. This persistence tends to be around .1 second. This causes the eye to perceive motion if the image is changing at a rate of greater than 10 frames per second (fps). It has been determined that smooth motion requires a frame rate > 15 fps.

RadarRadio

Infra

Red

700nm

400nm

X-RaysUltra

Violet

Color changes detected

by 8M cones

Gamma

Rays

Cosmic

WHITE PAPER (cont.)

.

V (Saturation)

Y (Luminance)

.

Analog Video

.

1-Wire Composite

ECG068/0798

.

Analog Composite Video

.

Analog video represents video information in frames consisting of fluctuating analog voltage

.

values. In early analog video systems individual video signals— brightness, sync, and color—

.

were all combined into one signal known as "composite" video. This composite signal can be

.

transmitted over a single wire. Compared to other forms of video, composite analog video is

.

lowest in quality. "Compositing" can result in color bleeding, low clarity and high generational

.

loss when reproduced.

.

Analog Component Video

.

The low quality of 1-wire composite video gave way to higher quality "component" video where

.

the signals are broken out into separate components. Two of the most popular component systems

.

are the Y/C 2-wire system and the RGB 3-wire system.

.

The Y/C system separates the brightness or luminance (Y) information from the color, or chroma

.

(C) information. This approach—called "S-Video"—is used in Hi-8 and Super VHS video

.

cameras. The RGB system separates the signal into three components—Red, Green, and Blue—

.

and is used in color CRT displays.

.

receive color signals, the color and brightness components were separated. Thus black and white

.

TVs could subtract out the chroma—hue and saturation—information of a color signal and color

.

TVs could display only the luma information received from a black and white transmission. This

.

enabled both types of TVs to peacefully coexist. In turns out that YUV signals can be

.

transformed into RGB signals and vice-versa by using simple formulas.

.

4

2-Wire Y/C

Component Video

Y (Luminance)

C (Chroma)

U (Hue)

3-Wire YUV

Component Video

R (Red)

G (Green)

B Blue)

3-Wire RGB

Component Video

Another approach to component video is to use Luminance (Y), Hue (U), and Saturation (V) as the three components. Hue describes the color's shade or tone, and saturation the "purity" or "colorfulness" of the color.

This approach dates back to the introduction of color TV. For color TVs to be backward compatible and for black and white TVs to be able to

WHITE PAPER (cont.)

.

4:4:4

4:2:2

4:1:1

= Luma (Y)

= Chroma (Cr, Cb)

.

Digital Video

.

A major disadvantage of analog video is that it tends to degrade from one generation to the next

.

when stored or reproduced. Another is that it often contains imperfections in the form of

.

ECG068/0798

.

"artifacts" such as "snow" in the picture due to noise and interference effects. In contrast to

.

analog video, digital video represents the video information as a series of digital numbers that can

.

be stored and transmitted error free without degrading from one generation to the next. Digital

.

video is generated by sampling and quantizing analog video signals. It may therefore be

.

composite (D2 standard) or component (D1 standard) depending on the analog source. Until

.

recently, digital video has been mostly stored on sequential tape because of the high capacity

.

requirements, but advances in magnetic and optical disk capacity and speed make it economically

.

feasible to store video on these media. To do this, analog video may be "captured" by digitizing it

.

with a capture card and storing it as digital video on a PC's hard drive.

.

This makes it possible to more easily retrieve it, search through it, or edit it. Recently, new digital

.

camcorders have emerged that store the video in digital form directly in the camcorder—usually

.

on tape, but sometimes on a disk in the camcorder itself. Digital video from these sources may go

.

directly to the hard drive of a PC by using an appropriate interface card.

.

The quality of digital video may be judged based on three main factors:

.

1. Frame Rate—The number of still pictures displayed per second to give the viewer perception

.

5

of motion. The National Television Standards Committee (NTSC) standard for full motion video is 30 frames per second (fps)—actually 29.97 fps)—where each frame is made up of odd and even fields, hence 30 fps = 60 fields per second. By comparison, film is 24 fps

2. Color Depth—The number of bits per pixel for representing color information. For

example, 24-bits can represent 16.7 million colors, 16-bits around 65,535 colors, or 8-bits only 256 colors.

3. Frame Resolution—Typically expressed as the width and height in pixels. For example, a

full screen PC display is 640x480; a quarter screen is 320x240, a one-eighth or "thumbnail" is160x120.

Digital Video Formats

To mimic the eye's perception of color, computer monitors display color information about each pixel on the screen using the RGB (Red, Green, Blue) format. Digital video, however, often uses a format known as YCrCb, where Y represents a pixel's brightness, or "luma", and Cr represents the color difference Red - Y, and Cb represents the color difference Blue - Y. By subtracting out the luminance Cr and Cb represent “pure” color. Together CrCb are referred to as "chroma".

WHITE PAPER (cont.)

.

these pipes

The '4' in the above descriptions indicates that luma is sampled at 4 times the basic 3.375 MHz

.

frequency and the '1' and '2' indicates that the chroma is sub-sampled at 1 or 2 times the basic

.

frequency. This approach to storing and transmitting video has the advantage of enabling file

.

ECG068/0798

.

size reduction without any noticeable impact to the human eye on picture quality. Since, the eye

.

detects subtle variations in brightness easier than differences in color, more bits are typically used

.

to represent brightness with fewer bits representing color information. In this scheme each video

.

pixel has its own luma, or Y, value, but groups of pixels may share CrCb chroma values. Even

.

though some color information is lost, it is not noticeable to the human eye. Depending on the

.

format used, this conversion can result in a 1/3 to 1/2 reduction in file size. The color bits

.

required per pixel for each of these formats is 24 (4:4:4), 16 (4:2:2), and 12 (4:1:1).

.

Network Delivery Challenges

.

The bandwidth required by video is simply too great to squeeze through narrow data pipes. For

.

example, full-screen/full-motion video can require a data rate of 216 MegaBits per second

.

(Mbps). This far exceeds the highest data rate achievable through most networks or across the

.

bus of older PCs. Until recently, the only practical ways to get video on a PC were to play video

.

from a CD-ROM or to download a very large file across the network for playback at the user's

.

desktop. Neither of these approaches is acceptable for delivery of content across a network.

.

The Bandwidth Problem

.

The scope of this problem can be seen by looking at the following illustration of available

.

bandwidth for several methods of data delivery.

.

As can be seen from this illustration, even a high bandwidth Ethernet LAN connection cannot

.

handle the bandwidth of raw uncompressed full-screen/full-motion video. A substantial amount

.

of video data compression is necessary.

.

Successfully delivering digital video over networks can involve processing the video using three

.

basic methods:

.

1. Scaling the video to smaller window sizes. This is especially important for low bandwidth

.

6

access networks such as the Internet, where many clients have modem access.

Technology Throughput

Fast Ethernet

Ethernet

Cable Modem

ADSL

1x CD-ROM

Dual channel ISDN

Single channel ISDN

High speed modem

Standard modem

100Mbps 10Mbps 8Mbps 6Mbps

1.2Mbps 128Kbps 64Kbps 56Kbps

28.8Kbps

Uncompressed

video at 216Mbps

and above, won’t

squeeze through

WHITE PAPER (cont.)

.

2. Compressing the video using lossy compression techniques. This is generally needed for

.

ECG068/0798

.

7

almost all networks because of the high bandwidth requirements of uncompressed video.

3. Streaming the video using data packets over the network. Small video files may be

downloaded and played, but there is a tendency to stream larger video content for faster viewing.

Scaling

While converting from the RGB color space to a subsampled YCrCb color space helps reduce file size, it is only a 1/3 to 1/2 reduction which is not nearly enough. Techniques to lower this further involve scaling one or more of the three factors mentioned above: frame rate, color depth, and frame resolution. For example, scaling the frame resolution results in different size windows for showing the video on the screen.

1/4 Screen 1/8 ScreenFull Screen

Further scaling of all three parameters can dramatically reduce the video rate as can be seen in the following diagram.

Frame Rate: 30 fps (Full Motion) Resolution: 640x480 (Full Screen) Color Depth: 24-bit (True Color)

Data Rate = (640 X 480 pixels)* (3 bytes/pixel)*(30 fps)/(1024000 bytes/megabyte)*8 bits/byte = 216 Megabits per second

Even though the above scaling represents over a 10:1 reduction in data rate at the expense of size and video quality, it is still not enough for most network delivery. For example, a 10BaseT Ethernet network supports data rates of 10 MegaBits/sec. This is not enough bandwidth to deliver even one video stream at the above scaled data rate. Further scaling can be done. For example, the video can be scaled to a "thumbnail" size video at a few frames per second with 8-bit color—but this is poor in quality and still does not accomplish the data rate reduction necessary to

x 1/2 x 1/4 x 2/3

Frame Rate: 15 fps Resolution: 320x240 (Quarter Screen) Color Depth: 16-bit

Data Rate = (320 X 240 pixels)* (2 bytes/pixel)*(15 fps)/(1024000 bytes/megabyte)*8 bits per byte = 18 Megabits per second

WHITE PAPER (cont.)

.

100111001100010

Streaming Files

deliver very many streams for most network delivery. To achieve further reduction in data rate,

.

video compression is needed.

.

ECG068/0798

.

Compressing—Codecs

.

Different algorithms and techniques known as "codecs" have been developed for compressing

.

video signals. Video compression techniques take advantage of the fact that most information

.

remains the same from frame to frame. For example, in a talking head video, most of the

.

background scene typically remains the same while the facial expressions and other gestures

.

change. Taking advantage of this enables the video information to be represented by a "key

.

frame" with "delta" frames containing the changes between the frames. This is typically called

.

"interframe" compression. In addition, individual frames may be compressed using lossy

.

algorithms similar to JPEG photo-image compression. An example of this is the conversion from

.

RGB to the Y/C color space described above where some color information is lost. This type of

.

compression is referred to as "intraframe" compression. Combining these two techniques can

.

result in up to 200:1 compression. This compression is achieved through the use of a “codec”—an

.

encoder/decoder pair, depicted as follows.

.

Video In

.

8

AVI

QuickTime

Files

A "Codec" is a combination of an EnCoder and a decoder

Codecs vary depending on their purpose, for example, wide bandwidth vs. narrow bandwidth or CD-ROM vs. network streaming. Encoders generally accept file types such as Audio/Video Interleave (AVI) and convert them into proprietary streaming formats for storage or transmission to the decoder. Multiple files may be produced corresponding to the various bit rates supported by the codec. A codec may also be asymmetric or symmetric depending on whether it takes longer to encode than decode. Some codecs are very compute intensive on the encode side and are used primarily for creating content once that will be played many times. Symmetric codecs, on the other hand, are often used in real-time applications such as live broadcasts. A number of codecs have been developed specifically for CD-ROMs while others have been developed specifically for streaming video.

Encoder Decoder

CD-ROM Codecs

Cinepak

TrueMotionS

Smacker

Video 1

Power!VideoPro

Transmit

Store

Proprietary

Indeo

Codec Types

MPEG

Video Out

To Display

Proprietary

Streaming Codecs

Vxtreme

ClearVideo

VDOLive

Vivo

RealVideo

TrueStream

Xing

WHITE PAPER (cont.)

.

Video Codec Standards

.

H.261

ECG068/0798

.

The H.261 video-only codec standard was created by the ITU in 1990 for global video phone and

.

video conferencing applications over ISDN. It was designed for low bit rates assuming limited

.

motion as is typical with videophone applications. It was also assumed that ISDN would be

.

deployed worldwide. Since each ISDN B Channel has a data rate of 64 Kbps, H.261 is also

.

sometimes referred to as "Px64" where P can take integer values from 1 to 30. For compatibility

.

between different TV systems—NTSC, PAL, SECAM—a Common Intermediate Format (CIF)

.

was defined that will work across displays for all of these systems. CIF and Quarter-CIF

.

resolution are defined as:

.

H.261 frame rates can be 7.5, 10, 15, or 30 fps.

.

H.261 has been the most widely implemented video conferencing standard in North America,

.

Europe, and Japan, and formed the starting point for the development of the MPEG-1 standard

.

described below.

.

H.263

.

H.263 was developed by the ITU in 1994 as an enhancement to H.261 for even lower bit rate

.

applications. It is intended to support videophone applications using the newer generation of

.

PSTN modems at 28.8 Kbps and above. It benefits from the experience gained on the MPEG-1

.

standard. It supports five picture formats:

.

Bitrates range from 8 Kbps to 1.5 Mbps. H.263 is the starting basis for MPEG-4 discussed below.

.

JPEG and MJPEG

.

JPEG stands for "Joint Photographic Experts Group." This group developed a compression

.

standard for 24-bit "true-color" photographic images. JPEG works by first converting the image

.

from an RGB format to a YCrCb format described above to reduce the file size to 1/3 or 1/2 of its

.

original size. It then applies a sophisticated algorithm to 8x8 blocks of pixels to round off and

.

quantize changes in luminance and color based on the properties of the human eye that detects

.

subtle changes in luminance more than in color. This lossy compression technique has

.

compression ratios in the range of 2-30.

.

9

H.261 Resolutions

Format Resolution

QCIF: 176 x 144

CIF: 352 x 288

H.263 Resolutions

Format Resolution

Sub-QCIF: 128 x 96

QCIF: 176 x 144

CIF: 352 x 288

4CIF: 704 x 576

16CIF 1408 x 1152