Media Server Technology: Karl Paulsen
Fibre Channel Switched Architectures
The power of high-speed interconnections for video servers continues
to shape the architecture of modern network-based storage systems.
As the use and deployment of these systems grow, of paramount issue
is the protection of the stored assets from disruptive activities,
an important part of meeting the needs of a continuous full time
operation. A variety of techniques are available for protecting
the components in a network-based storage architecture. Some of
these approaches grew out of the natural course of storage expansion.
Other methodologies came from the inherent need to increase system
bandwidth, which grew from a few hundred megabits per second in
the early SCSI only-based arrays, to thousands of megabits per second
in the one- and two-gigabit Ethernet domain.
As the requirement for increased storage is met by adding groups
of higher capacity disk arrays, the storage network architecture
must also be increased to direct and handle the demands of the server
engines themselves. The techniques employed in broadcast video servers
are similar to those used in the data-centric arrays of storage
systems, but with one critical, and often reminded upon, requirement
-- the system must maintain the capability to deliver large continuous
blocks of isochronous data to real time decoders without interruption
or corruption. This is the key, and single greatest difference between
streaming data for wide area networks and the delivery (and recording)
of continuous moving video images from a storage platform.
Video server manufacturers approach storage, storage networking,
and the encoding or decoding of the media in similar fashions. Setting
aside for the moment the issues of file formats and interchange,
this installment will focus mainly on the methods and means of connection
between the server chassis or engine and the storage subsystems.
Using carefully qualified openly available components, the video
storage systems which comprise these ever-increasing in size (and
use) video servers use similar interconnection methods, but with
specialized applications that allow the manufacturer to tailor the
drive arrays to the specific needs or architecture of the system
they are delivering.
PROTECTING YOUR ASSETS
Facilities migrating from depending on videotape for their program
content and commercial playback to a fully integrated video server
architecture give considerable thought to the protection of their
assets. In the non-video-disk days of videotape, broadcasters generally
used a second or third method to protect their pre-recorded programming
for the air chain; they would double-record the program and even
dual-roll the playback from two transports at the actual time of
air. Provided the media wasn't damaged, videotape transports permitted
the physical exchange of media between them, a practical and relatively
simple means for protection. Mission-critical time-delayed material
was generally recorded on a similar backup transport, or in some
cases on a secondary 3/4-inch or VHS transport. The costs for a
second transport were reasonable, and in most cases justifiable.
Comparatively, video server protection schemes are built on a different
model. Fundamentally, it is impractical to physically change the
media between a set of arrays or move the disk drives from one server
platform to another when something fails. Furthermore, moving a
single hard disk drive in an array from one storage chassis to another
is a worthless exercise due to the general incompatibility of the
file structures on each set of independent arrays. Therefore, as
the principles of networking come into play in the video server
system, you realize the benefits of redundancy and protection.
The first and most basic form of protection in a storage system
is handled by the Redundant Arrays of Independent (or Inexpensive)
Disks configuration, whether that be a RAID 1, RAID 3 or RAID 5
implementation. In RAID systems, the overall data protection is
achieved by striping data over multiple storage drives and mapping
the data protection (the parity data) among and between them. The
loss of one drive, or in some RAID configurations even two drives,
becomes less of a problem than the loss of a single transport or
the destruction of a sole videotape master. In addition, given these
basic principles of RAID for protection in the media server, other
advantages are achieved, including increased bandwidth, greater
system throughput and minimal requirements for immediate human intervention.
The converse is also true: As bandwidth increases, so does protection.
To increase bandwidth, many systems parameters are extended. First,
we must add faster drives in more efficient storage arrays. Second,
we spread that data over multiple arrays to increase access while
simultaneously increasing storage capacity. Third, we provide a
means to exchange that data between both data stores and server
engines at high speed transfer rates. Fourth, we add protective
measures to handle instances where a key element fails and the balance
of the system fails over -- or a secondary system automatically
intervenes and provides a means for recovery.
With the introduction of Gigabit Ethernet and Fibre Channel switches,
that third element -- "a means to exchange data between stores and
server engines" -- has become a smoother and more universally accepted
process in implementation. Again, it is the principles of networking
which are playing an important and significant role in the development
of the broadcast video server for multiple purposes and large-scale
system architectures.
Fibre Channel, a channel/network standard, is now being utilized
by nearly all broadcast video server manufacturers. Fibre Channel
has extended beyond that of a single arbitrated loop, which moved
data at faster than real time between one or more server chassis.
Fibre Channel, and Gigabit Ethernet are both opening the neck of
the bottle to allow for more data flow and better control of the
data. As video server systems scale from medium to large systems,
many are now utilizing 1 and 2 Gbps Fibre Channel switches as the
data channel manager that connects between multiple server engines
and their associated storage arrays. To manage the growing speed
and amount of data flow between devices, one or more Fibre Channel
switches may be employed. Using a managed switching system improves
both the bandwidth management and provides for a higher level of
system protection.
Once thought of as expensive, difficult to configure and manage,
the Fibre Channel switch is rapidly becoming a mainstay element
of the broadcast video server architecture.
For scalability of storage and system bandwidth, the FC-switched
architecture provided much better overall performance, especially
when 2 Gbps Fibre Channel products are utilized.
 |
| Fig. 1: Switched Fibre Channel Protected Server Architecture |
Fig. 1 depicts a multiserver, multi-I/O design utilizing a scalable
storage array structure interconnected through a pair of redundantly
configured Fibre Channel switches. The components consist of two
server engines (identified as A and B), which can be operated in
either a mirrored or independent structure, a set of storage arrays
(F), and a pair of Fibre Channel switches (C and D).
When the facility operates with a protected set of video outputs,
then the outputs of server B will be configured to match that of
server A. In detail, a primary "A" playout channel and a secondary
"B" playout channel both playback the same data to two different
decoder outputs (in two separate chassis) whose baseband outputs
feed digital video routers or master control switchers. In operation,
two decoders in two separate chassis are working in parallel, each
pulling the same data (video clips) from the same set of common
storage arrays (shown as F). The "A" playout channel is synchronized
with "B" playout, allowing an instantaneous switch to the opposite
output from the other server engine should the primary server output
fail. In Fig.1, only a single common, but well-protected storage
system is used.
Operations that do not require a protected playout could also utilize
the same decoder/server arrangement, but instead of parallel, synchronized
operations in two chassis, there are now twice as many outputs available
for the playout of different video streams to different program
channels.
Each of the server engines (A and B) have a set of dual FC-ports,
which are connected to the FC-switches, shown at points C and D.
For protection purposes, should one FC-switch fail or a FC-channel
data port from a server fail, there is a cross connection to the
other switch (shown as E) which maintains a seamless "transfer of
operations" between the storage systems, server chassis or FC-switches.
The "cross-over" channel between the two FC-switches serves two
purposes: First, it acts as a protective measure should either of
the switches fail; and second, it is the means of connection between
the two sets of server engines to the data storage arrays via redundant
paths.
Each storage array also has two sets of Fibre Channel I/O-ports
that connect to each of the FC-switches as shown. Should a port
interface fail on any storage array, the opposite path immediately
picks up the data and routes it to or from the respective FC-switches
and on to the server engines. Management software regulates the
flow of data and actions of the FC-switches in various demand or
protection modes.
PEACE OF MIND
Providing this level of switched management produces an extra level
of reliability, and it provides a path for serviceability while
minimizing down time. Individual system components can be removed
and/or replaced without unduly harming the overall operations. This
concept can also be employed in complex news editing and program
ingest/playout systems. Such extensions include another complete
"mirror" of this system interconnected via FC-switches, redundant
storage array, and streaming to remote location via a gateway for
disaster recovery.
As broadcast video server systems are proposed for integration
into the facility, their structure must be designed to meet the
workflow methods of the operations in which the systems will be
used. In order to achieve maximum performance from the system, several
operating parameters must be optimized. The planning of these larger-scale
systems is complex and specialized, and generally specific to the
products and techniques used by the broadcast video server manufacturer.
Optimizing the performance and management of these advanced 2 Gbps
storage area network fabrics is just one of the issues associated
with the construction and integration of the modern broadcast video
server environment. Other issues that need to be considered as these
high-speed networking components are configured for the applications
in which they are placed include: reducing congestion, increasing
data availability, switched system aggregation (referred to as trunking),
and the maximization of data transfer rates.
In a later issue, this column will further explore the rationale
behind the selection and uses of the Fibre Channel switch; including
how the WAN is being used for remote location and disaster recovery.
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.
|
