Media Server Technology: Karl Paulsen
Essential Elements for The Video Server Facility
Part 1
Scaling a broadcast, newsroom or production facility toward a
complete video server environment involves numerous networking components
beyond the video server's encoder/decoder, its storage and associated
application software. As video material management moves from videotape
and more toward a tapeless production environment, the architectures
of these systems begin to take on the same dimensions as baseband
video-only systems-but without the numerous dedicated VTRs, video
switchers and physical videotape libraries we've grown accustomed
to over the previous two decades of digital video implementation.
Fundamentally, the essential building blocks of the compressed
digital/tapeless environment appear similar: There is a means to
transfer field material into the edit environment, then editing
is done on that material, and finally the finished products are
sent for playout or distribution. But that is where the similarity
from a component standpoint ends. Since the inception of nonlinear
editing (NLE) in the early 1980s, which has signaled the eventual
departure from the serial processes of videotape production, the
workflow and time necessary to complete a production element has
decreased dramatically. In turn, the cost to produce those production
elements has decreased as well. These two components, time and cost,
continue to fuel the development of the products and the deployment
of the nonlinear, entirely server-based facility.
BUILDING BLOCKS
In this installment we will look at the building block components
used to build a complete facility including ingest, server playout,
production/editorial, archive and off-site asset protection for
disaster recovery purposes. The example shown in Fig.
1 is not an artificial concept. Indeed this architecture is
typical of systems being deployed by all the major video server
manufacturers, albeit some take different approaches to the networking
concepts, and their video I/O structure may differ depending upon
the server engine's internal architecture.
As expressed in my previous column, protection of the asset and
redundancy in systems are the major keys to a continued successful
operation. In the accompanying diagram, nearly every portion of
the system has at least one alternative path for both the data transfers,
the system management and the operational processes. Furthermore,
all stored data can be shared between all the elements of the workflow
processes from ingest and playout, through editorial and network
interconnection for remote purposes. What differs between the various
manufacturers' implementations are the software applications and
the degree (and methods) of data transfer between various operational
work areas. For those instances, I encourage you to contact specific
manufacturers for their individual representations of features and
functionality, as each facility most likely will require its own
special configurations and components, depending upon its operational
model.
The "system" architecture shown in the example is comprised of
PROGRAM EDITING, PRODUCTION EDITING, INGEST AREAS 1&2, AIR PLAYOUT
with QC (quality control), STORAGE, ARCHIVE, and a NETWORK INTERCONNECT
component. Each system, with the exception of STORAGE and the FABRIC
SWITCHES-which are the common elements in the overall system-can
function independently and autonomously from one and other.
STORAGE
To begin, we will first look at the most central element of the
system, the storage component. In our example, all the compressed
digital essence (video, audio and metadata) will be kept in a common
"pool" shown by the STORAGE portion of the diagram. Each storage
array has a dual-port Fibre Channel interface that moves data to
and from the rest of the system. Data flows out both ports, in either
direction (shown as the left or right side of the storage array
diagram) to a pair of Fibre Channel fabric switches, configured
in a redundant mode such that either path from the storage arrays
can sustain the full bandwidth of the system as designed. It is
these DUAL FABRIC SWITCHES (see "A") that act as the traffic control
mechanism between all the system components, managing the data transfer
between ingest and playout servers, edit stations, the archive server
and the two GATEWAY INTERCONNECTS ("C") that bridge this local system
to other external systems. The GATEWAYS allow the system to be integrated
with other server systems using Gigabit Ethernet, over a wide area
network or to another video server-based systems, such as a newsroom
editing and computer network.
A local or NEAR LINE ARCHIVE is bridged into the system through
the fabric via an ARCHIVE SERVER ("B") that manages and buffers
the transfer of compressed data between server and data tape or
DVD storage. The archive server may further employ additional third-party
software or a second data server to process the activities specifically
associated with the digital tape or DVD archive.
The data storage methods and arrays are especially particular
to the actual video server manufacturer. Most storage arrays and
associated Fibre Channel switching systems are selected by the manufacturer
to meet certain specific requirements specific to that individual
manufacturer's design. For example, in some servers the data may
be spread (or striped) across multiple drive array chassis, allowing
for both RAID protection and increased bandwidth when multiple data
transfer requests are issued from the playout or editing station
operations. Other server designs will segregate the arrays into
groups and control access to those areas via software and the Fibre
Channel switches.
INGEST AND PLAYOUT
From a workflow perspective, the example diagram breaks the ingest
and playout areas into segments across the overall network. For
example, INGEST & PLAYOUT AREA #1 is intended to record feeds such
as live local production or videotape via the central house video
router. Typically these sources are controlled from the central
equipment room or satellite record center and continusouly monitored
from this same area. Hence, there are inputs to server A and server
B, but a single output from server B, which functions as the "QC"
station for this activity.
INGEST AREA #2 would typically become the satellite, network delay
or turnaround area. Server C and server D only have input encoders,
and the servers typically would never be involved with playout,
either for QC or air playback purposes. The number of inputs or
outputs is determined by the manufacturer or type of video server
product selected. Note that some server manufacturers actually have
bidirectional channels-ones that can either encode (ingest) or decode
(playback) or feature internal baseband digital video routers that
steer the signal into internal codecs; creating an additional level
of feature sets that should be considered depending upon the workflow
and operational needs of the system.
Integrated tapeless editing systems (such as depicted in the PROGRAM
and PRODUCTION EDITING STATIONS) are migrating from the baseband
video I/O model associated with the dedicated or standalone NLE
system. Today's systems are entirely operated at the compressed
digital level, meaning that no transcoding occurs from MPEG-2 to
component digital (ITU-R 601) at the workstation location. Depending
upon the system deployed, the original baseband or compressed video
media is cached into the central STORAGE system via a separate server
engine. Time-line editing occurs at the EDIT STATIONS using data
that is stored only on the STORAGE arrays.
Not having to transfer that "videodata" to each workstation for
editing or viewing reduces editing time and the amount of hardware,
as well as storage, required on the system. Essentially, the edit
workstation uses either proxies of the compressed digital video
or creates only a set of pointers to the files that are kept on
the storage array. Only the effects (dissolves, keys, etc.,) are
rendered at the actual workstation, then those clip segments are
transferred back to the central store-keeping both the original
material and all the final material on the central storage arrays.
Once a segment or the entire cut is completed, the user can either
create a final leveled (that is rendered, separate clip) on the
central store for playback or archive-again leaving the original
material untouched and unaltered.
These larger systems now permit a great deal of activity to happen
concurrently. However, with multiple edit sessions happening simultaneously,
and continual caching of material to the stores ongoing, a lot of
bandwidth can be consumed on the "network." High-speed networking,
both over Fibre Channel and Gigabit Ethernet, is now used in a variety
of combinations depending upon what functions are occurring in the
system. In addition, a separate 10/100 base-T Ethernet control system
(not diagrammed for simplicity) overlays the high-speed data network
for the carriage of control and other metadata between elements
of the system. This 10/100 base-T Ethernet data network is used
as a signaling system for edit decision list management, monitoring
the health of the network and determining or controlling the bandwidth
management schemes during peak data transfer activities. These components
can also be integrated with a browse or proxy system for offline
desktop preview or rough-cut editing.
SWITCHING
The various component elements of the total system are spread
across one or more pairs of fabric switches for at least two purposes
(note only one pair is depicted in this diagram). First, for the
management of the workflow process, some servers may be grouped
with certain edit stations to balance the amount of data flow between
elements through the switch. Second, in the event of a fabric switch
failure, the entire system is not crippled because although two
sets of functions may be temporarily disabled, other workstations
could pick up the load because all the data essence material is
still accessible from the central STORAGE arrays.
Extremely complex or very large-scale systems may use multiple
sets of fabric switches in an arrangement that does not hinder operational
performance, should any single network component fail. Multiple
switches allow the system administrators to reconfigure the digital
video network for other purposes. This level of system discussion
is beyond the scope of this article, but it is not uncommon to find
extensions of this diagram applied to extremely large-scale implementation
at news organizations such as CNN or others.
The dedicated QC & PLAYBACK areas and the balance of this diagram
are somewhat self-explanatory at the block level. In the concluding
portion of this installment, we will look at how the ARCHIVE and
the GATEWAY components are further extended for off-site mirroring,
wide array network distribution, streaming video, backup and other
campus- or facility-wide implementation.
Karl Paulsen is vice president of engineering at AZCAR (www.azcar.com)
and the author of the book "Video and Media Servers: Technology
and Applications-2nd Edition" (published by Focal Press). Contact
him via e-mail at karl.paulsen@azcar.com.
|
