Video on Demand
Movies! Karaoke! News! They're all available at the touch of a remote
through Video on Demand
In Brief:
A trial of futuristic technology that permits viewers to select movies,
karaoke and business programs, including news, at will, and to view them as they
would their own VCR tapes, is under way in Tokyo. IBM Japan has integrated a
variety of technologies developed by Research to create a scalable video server
complex for interactive television. These include systems that distribute video
among several disks, allow parallel computers to store and retrieve data
simultaneously, request data delivery at specific rates, and serve as the
interface between the parallel computers and the cable delivery system.
New technologies that are critically important to consumers rarely emerge
full-blown. They usually build up in stepwise fashion, as technological advances
make individual components available. The Internet, for example, started out as
a specialized mode of communication for the academic research community. Only
over the past five years has it opened up to a broader community, thereby
stimulating a succession of improvements that have made it more effective and
more user-friendly.
That type of development is now occurring in a communications technology
geared even more to the general public. Video-on-demand (VoD) is a futuristic
cable TV system that lets users select and view popular movies and other
programs whenever they want. The concept differs significantly from conventional
cable TV. Not only does it make programming from an on-screen menu instantly
available to subscribers whenever they want it, but it is also interactive. Like
a VCR, it lets subscribers pause the program while they get a snack, replay
parts they missed or liked and fast-forward to new content.
At present, VoD is some years from large-scale commercial reality. To reach
it, Research teams are making technological advances that, in the process, are
forming the basis of new products. In addition, VoD and interactive TV are
already under development for hotels, businesses, schools and cruise ships that
have specialized cable or fiber installed.
"Although these applications are not yet cost-effective for widespread
deployment, it is only a matter of time before the underlying chip and access
technologies will permit such interactivity," says Jurij Paraszczak, senior
manager, telecommunication systems, at IBMs Thomas J. Watson Research Center. "This
will produce profound changes in the way that we think of todays computers, TVs
and entertainment systems and will allow IBM to capitalize on its strengths in
networked delivery of content and advanced chip integration technology."
The most ambitious trial of the technology is occurring in hundreds of homes
in a new island community in Tokyo harbor called Tokyo Teleport Town. There,
residents are undertaking a pioneering interactive trial of VoD as a showcase
for advanced communications technology. The one-year trial, which started in
June 1996 under the auspices of the Tokyo Metropolitan Government (TMG), has the
goal of tracking the feasibility of VoD systems for broader application in Japan
and the rest of the world. In addition to households, the representative
project includes connections to offices inside and outside the region and to a
demonstration center at the Tokyo Telecom Center Building and International
Exhibition Center.
TMG selected IBM Japan's bid to integrate and oversee the trial, in part
because of its earlier success in bringing interactive VoD to classrooms in
Okazaki, Japan. Working with several partners including JVC, Panasonic,
Pioneer, Sony and Sumitomo IBM Japan took the new project from concept to
reality in two years, delivering Hollywood to the first hundred area residences
by the dedication on June 1, 1996. In the process, IBM Japan not only designed
the overall framework of the entire system but also developed the application
platform for interactive TV based on international standards.
In their effort to integrate and deploy an end-to-end interactive VoD
system, IBM Japan drew on the resources of IBM Research groups, as well as from
IBM Brazil. The reason: interactive VoD makes demands on video delivery that are
very different from those of conventional CATV broadcasting, including
pay-per-view services.
"Our intent was to integrate core technologies developed in cooperation
with IBM Research," says Hideto Morishita, the TMG project manager at IBM
Japan. Visits to IBM laboratories in the U.S. convinced him that a competitive
solution to TMGs needs could be assembled from the advanced digital video
technologies that he saw under development. "Most of the key components
were already there," he notes.
The basic infrastructure
Heres what Morishita had in mind: movies, karaoke, and business programming
from film, CDs, videotape and other media would be digitized, compressed and
stored on conventional computer disk drives. A video server capable of providing
the stored video in real time at rates necessary to serve the needs of hundreds
of households would send the video demanded by each viewer over a special cable
system. In each clients home, a set-top device much like a small computer would
decompress the signals from the cable and direct them to a conventional TV for
viewing. The set-top box would also process viewers control requests (select
program, start, stop, for example) and send those signals back along the cable
to the video server for appropriate action.
The scenario took advantage of the modern cable infrastructure already
installed in the community: a combination of the coaxial cable used in most CATV
systems and the fiber optical cable usually associated with high-volume
telephone links. This hybrid fiber/coaxial cable (HFC), IBM decided, could carry
multiple video streams from the VoD server to the set-top boxes and still have
sufficient bandwidth to carry the commercial broadcast channels of Japan's Tokyo
Teleport Center.
However, compatibility loomed as a problem. "Our video server
technology needed to interface with the HFC cable TV network, and that was a
missing link in our solution when we began investigation in 1994," says
Isamu Terao, manager of digital network business promotion at Cross
Industry-Network Computing Promotion, IBM Japan.
TMG wanted to provide a VoD selection of 30 movies, 500 karaoke songs and
180 business topics, representing about 200 hours of digital video. Even when
compressed, that adds up to 720 gigabytes of data and requires a server capable
of handling such a large amount of storage. Data handling requirements were
large, too. An estimate that one-fifth of the 500 households in the Teleport
Town would access VoD at any given time indicated that the server would need to
deliver 100 simultaneous video streams. To provide high- quality TV images and
sound, each stream would have to supply uninterrupted video at the rate of 6
megabits per second a total throughput of 600 megabits per second.
Seeking a suitable server
Long before IBM Japan and TMG began talking, IBM Research groups were
crafting the technologies needed to make a VoD server possible. Scientists at
the Almaden Research Center were constructing a new kind of file server to
handle the unique demands of video. At the Thomas J. Watson Research Center,
work on a "digital pump" tackled another problem that the server and
disk storage could not handle: providing the interface that moves the video onto
HFC cable and then to a TV set-top box.
The search for a server that could handle the special needs of VoD began at
Almaden in 1992 under Roger Haskin, manager of parallel file systems. "We
began work on a new kind of server for video to satisfy our own curiosity about
feasibility," says Haskin. "For video, you have to continuously read,
or stream, data off a disk drive in real time. Most conventional data servers
slow down when demand for data exceeds the output level for which they were
designed. While that is a graceful solution for users who may have to wait only
a fraction of a second longer to perform a database transaction, it is an
unacceptable situation for video."
Almaden's early solution, called Shark, was used in 1993 in one of IBM's
first VoD trials, in collaboration with Bell Atlantic. A system based on an IBM
RS/6000(TM) workstation delivered 50 simultaneous video streams at 1.5 megabits
per second each, thereby demonstrating the feasibility of VoD and the new server
strategy. A subsequent trial at Hong Kong Telecom in 1995 sent out 150
simultaneous video streams, again at 1.5 megabits per second. This involved two
IBM RS/6000 SP® scalable parallel computers and, for the first time, a
load-balancing technique known as wide striping (see
below) across multiple
disks. With the advent of wide striping, the file system was appropriately
renamed Tiger Shark.
At about that time, Haskin's group and Tiger Shark were drafted to play
critical roles in the TMG project. "High data rates and the large number of
streams forced us to use an IBM SP, a much more powerful processor,"says
Haskin. "An SP is actually several RS/6000 processors teamed together to
split up the work into parallel streams for greater throughput. For the TMG
project, we used an SP with 12 processors, or nodes, to achieve the throughput
we needed."
Equal access
The team decided to store just two copies of each movie, karaoke song and
business topic on a cluster of 160 disk drives, totaling 750 gigabytes of
storage. To give each node of the SP access to the data, Watson researchers
devised a shared disk approach. "We modified the virtual shared disk (VSD)
technology we had developed for parallel databases to create a real-time
version, which runs under Tiger Shark," says Dan Dias, manager of parallel
commercial systems at Watson, who worked on the project with Haskin.
The new approach is like physically connecting all of the drives to all of
the SP nodes, notes Watson research staff member Rajat Mukherjee, "except
we do it in software and in real-time by making all the disks accessible to all
the nodes."
In order to maintain a consistent view of the content on multiple SP nodes,
Tiger Shark uses a distributed lock manager developed for another file system
called Calypso, developed at Watson. Calypso is also used as a directory service
in the TMG system, whereby it translates a symbolic file name into an internal
identifier that can be used to obtain data blocks of the file.
But with only two copies of a movie available, the question arises of how
the system can meet the demand particularly when hundreds of users opt to watch
a highly popular movie like True Lies. Thats when the Tiger Shark
technology proves its worth, by breaking up the movie into small segments that
it distributes over multiple disks.
The SP server with Tiger Shark technology scales up well, according to
Haskin. Initially, with about 100 users, the server was streaming video at about
one-tenth of its maximum design rate of 750 megabytes per second. "By
comparison, a conventional file server will run out of gas at about 10 megabytes
per second," he says. "Were up to 500 users now and not seeing any
signs of video or audio breakup."
No Mirag
The TMG trial needed one more key component: a system that would order and
transmit the correct amount of video at the correct data rate over HFC, based on
viewers requests. Early in 1994, a team led by Ahmed N. Tantawy, then at Watson
and now director of video solutions technology at IBM's Telecommunications and
Media ISU in Somers, NY, prototyped what he called Mirag (and pronounced
mirage). This sits at the broadcast end of the CATV system and pushes video out
to viewers, using the industry standard MPEG-2 transport system specifications
in order to multiplex several digital video streams over each regular TV
channel.
Mirag technology is the basis of the MPEG Stream Connection (MSC) card. A
number of MSC cards reside on the cable side of each individual SP node each of
which can serve many users simultaneously and interfaces with the HFC cable. By
using the built-in switching capabilities of its PC or workstation environment,
the Mirag architecture overcomes the need for expensive switching technology.
By June 1995, a year before the project was to start in Japan, a prototype
that included MSC cards was running at Watson. According to Tantawy, the
prototype proved that Mirag was simpler to build, manage and scale up than other
technologies, and five times less expensive than the competition.
Mirag also provided a PC card for MPEG video decoding (called MVC) and
resource management software. The MVC card can be viewed as a digital cable box
on a card, and can be used to enable a PC for VoD service or as a testing and
monitoring device.
For the TMG project, IBM Japan carried out large-scale manufacture, testing
and integration of the Mirag cards. The unit also developed the network access
routine (NAR), a piece of software that provides an interface between the MSC
card and the SP network node. The Data Exporter, developed at the Haifa Research
Laboratory, acts as a "transmission" between Tiger Shark, the NAR and
Mirag, assuring a smooth data flow.
Control server
If the data pipeline from the real-time VSD (RVSD) over Tiger Shark, the
Data Exporter and Mirag are the heart of the video server, the Control Server
may be likened to its brain. The Control Server is the agent through which
applications request and control their videos. VCR-like commands give the
applications a convenient way of playing, pausing and seeking videos, just as
with a real VCR.
Ensuring that the video server meets its commitments for delivering data is
another task of the Control Server, developed by a group at Watson under Martin
G. Kienzle, manager of parallel multimedia servers. It processes viewers
requests sent to the VoD center from set-top boxes. "When a new request for
a movie comes in, the Control Server makes sure the system has enough available
capacity, then grants the request," says Kienzle. "The Control Server
will bar any request that, because of limited capacity, would degrade video
streams already running." An earlier version of this resource management
and control function was used in the Okazaki trial.
When starting a video stream, the Control Server selects a data path with
sufficient bandwidth, balancing the workload across all system resources. That
is particularly important for large systems involving many SP processing nodes.
When videos are loaded onto the video server the Control Server determines where
to place the video in the Tiger Shark file system. Using estimates of the videos
popularity, the Control Server ensures that it won't have to turn down many
requests for that video later.
Integration by cooperation
Without Researchs participation, the $30 million TMG project would surely
have taken longer to complete, in the view of Paraszczak. "Research
provided advanced technology that vaulted IBM into an arena where few have
succeeded and helped bring the project in on time and on budget," he
explains. Indeed, the scope of the project initially raised some concerns on the
part of the TMG, but it is pleased with the result. In the opinion of Sumihiro
Ueno, executive director of the Association of the Tokyo Multimedia System, "Only
IBM could have integrated such a large-scale system."
"The TMG system would not have been possible without advanced Research
technology, but the task of integrating that technology from Research sites
around the world was a logistical challenge," says Yoh Satoh, manager of
digital network technical support at Cross Industry and NC Promotion, IBM Japan.
"The need to work quickly and across multiple times zones added to the
pressure. In some case, however, the time differences between Japan and the
United States worked to our advantage. There were instances when, at the end of
the day, we would ask our colleagues at Watson to solve a particular problem,
and by the time we returned in the morning we would have our answer. The great
cooperation between Research and IBM Japan is something we can all be proud of."
VoD technology branches out
The efforts exerted to develop the technology used in the VoD system are
already providing extra dividends in new product offerings and unique customer
solutions outside Japan. For example, Multimedia File System (a.k.a. Tiger
Shark) is the basis for IBM Austins AIX¬ Multimedia Server, a single
RS/6000 video server for local area network PC and AIX clients that became
available in February 1996. Similarly, IBM technology developed for TMG helps
STET, the Italian telecommunications giant, build its Near Video-on-Demand
(NVoD) Digital Video Broadcast (DVB) system. IBM was the general contractor for
this project, that went into production on time in September, 1996. The TMG
solution has also been replicated in Kobe, Japan.
Within a decade, the ability to select and play favorite movies from an
online VoD archive will be widespread. By then, high bandwidth cable
infrastructure will be at almost every doorstep, and the cost of both set-top
boxes and server systems will be lower than it is now. Research is looking in
other directions, as well. The video server architecture was designed from the
outset to work with other delivery networks. This has paid off. The Research
components of the server, except Mirag, are the basis of two recently announced
IBM products: the Video Charger and the Media Streamer. The former is an
advanced server for intranets and the Internet that delivers videos to be shown
from Web browsers. The latter is aimed at the commercial video production and
broadcasting environment to support their transition from analog to digital
television.
Research work on video servers, therefore, not only served to demonstrate
the potential of the technology through actual customer trials, but has charted
the course for IBM's entry into digital video delivery on a very broad front.
Michael Sinclair is a freelance science writer based in North
Guilford,Connecticut.
More Information:
WIDE STRIPING: GETTING THE MOST FROM THE LEAST
Wide striping across multiple disks, the technology that forms the basis of
IBMs Tiger Shark, makes it possible for many people to view different parts of
the same movie at the same time, even though only one copy is stored on the disk
cluster. Heres how it works:
The movie is divided into many short segments and spread evenly across all
disks, a process called statistical multiplexing. For example, with 160 disks,
1/160th of the movie could be stored on disk 1, the second 1/160th on disk 2,
and so on until the entire movie is stored. Imagine that 100 people start
watching the same movie one evening.
At any specific time, most of the 100 will be viewing different scenes, so
the chances are that no disk will have to carry an excessive amount of the
workload and that plenty of bandwidth will be available. The system can even
satisfy two viewers who demand a movie at exactly the same time. In that
situation, it takes the same stream from each disk and sends it simultaneously
along two cable lines. However, that would break down without striping, because
of the high demand on individual disks.
Commercially available RAID (redundant array of inexpensive disk) subsystems
already stripe across several disks to help recovery from disk failure. Wide
striping across hundreds of disks is relatively new, and it has an entirely
different goal: load balancing that permits a system to handle a large volume of
viewers requests for a small reservoir of original programs.
Wide striping has one disadvantage: if a single disk crashes, it affects
every stream. The designers of Tiger Shark overcame that difficulty, by
incorporating replication strategies that automatically make requests to
alternate replica blocks if a disk dies. One approach is the use of two copies
of programs, rather than one, which are distributed in different ways across the
disks.
