Media Server Technology: Karl Paulsen
Connecting It All
When building any form of network, elements for consideration
include manageability, versatility, compatibility and cost effectiveness.
Data networks seldom emerge as complete replacements for existing
infrastructures; rather they are deployed as new or updated data
centers that are married into existing data networks, spanning either
work groups or entire enterprises.
For the broadcast or television system facility deploying
video server and storage systems, choices in hardware employed were
confined to a sole source – the video server manufacturer.
Typically, the vendors were charged with complete responsibility
for the selection of equipment, subsystems and applications software.
They would qualify all the equipment and only support those components
that were "packaged" into their nearly proprietary solution.
As technology advances, we’ve seen third-party
systems (such as tape archives and archive management software)
attached to video servers’ engines and disk storage systems.
As storage needs increased, these engines were interconnected over
additional third-party components (Fibre Channel switches) using
– again – another set of third-party subsystems (intermediary
buffers or gateways) comprised of conventional network servers in
both Unix (typically Sun Microsystems) or Windows (Wintel systems
on Dell, HP and others).
When browse capabilities, automation and media asset
management software is added, the user must turn to another set
of components, sometimes further separated from the video server
manufacturer’s core equipment.
ANOTHER DIMENSION
To move video server storage technologies to the another
dimension – one that follows models employed in the IT domain
– techniques must be added that not only allow for expanded
storage capabilities, but also allow for increases in bandwidth
and therefore overall performance – all the while maintaining
reliability and versatility.
Recalling that IT environments are virtually asynchronous
and video requires isochronous behaviors, maintaining sufficiently
high throughputs – so all channels can play-out (or record)
continuously – requires sophisticated designs and complex hardware
configurations.
For storage systems, one technique employed by the
video server manufacturers is the striping of the data across several
elements of one or more drive array subsystems. While the typical
practices of RAID are continued, the parallel synchronous movement
of large blocks of data is needed for video – and this is what
typically sets video server architectures apart from IT-based servers
and storage systems.
The number of inputs and outputs for the video server
systems – as well as the data rate for playback and encoding
(recording) – are key factors that are restricted by the overall
bandwidth of the system. For example, when a system needs playback
of five streams at 15 Mbps each, then the bandwidth requirements
must be at least 75 Mbps just to get data from the drives, through
the decoders and out of the system as video.
Transferring between storage systems (or server chassis)
must be completed at rates much higher than the playback rate –
typically 100 Mbps or more. This is bandwidth above and beyond the
play-out requirements, but taxes the overall system nonetheless.
During playback, continuous bandwidth integrity is
key. Without continuous, isochronous data from drives to decoders
– provided in sufficient quantity to each decoder – a
continuous stream is not possible.
Unlike software-based PC codecs (as in Real Video
or Microsoft Windows Media Player), video facilities simply cannot
wait for several seconds while a sufficient amount of data is buffered
(e.g., stored locally) before playback commences. Any buffering
must be continuous and uninterrupted, meaning the data must "stream"
off the drive array in a continuous and uninterrupted fashion.
When larger scale video server systems are deployed
– with dozens of playback and record channels all running different
media content (data) – system architectures must be expanded
in a nonlinear manner.
If additional functions, separate from record or playback,
are present – such as Fibre Channel transfers between other
servers or edit systems, or between tape or disk archives –
the requirements and expectations of the "video server"
grow tremendously.
Once peak system bandwidth is reached, performance
drops and unpredictable things begin to happen. (This is a principle
reason why video server manufacturers prefer to qualify and provide
all the components in a system.) Building the video server’s
storage architecture is therefore significantly different than building
an IT network.
Still, we’ve seen that the foundations of video
server storage have their roots in the IT world; this brings us
to the discussion of advances in storage architectures that, once
again, show the IT domain leading the charge.
MIRROR, MIRROR
Early nonlinear editing systems and broadcast servers
deployed SCSI interfaces between a rather small subset of disk drives.
Broadcast server storage was usually configured in a RAID 3 or RAID
5 arrangement, and often replicated 100 percent as a totally mirrored
system.
As storage needs increased – and Fibre Channel
for IT environments became mature – some of the video server
manufacturers found a need for interconnecting between ingest servers,
play-out servers and online libraries. Fibre Channel Arbitrated
Loop (FC-AL) was adopted as a principle storage transfer topology
between these varieties of systems and subsystems.
As more Fibre Channel products – including FC-based
hard drives – became available, the video server manufacturer
moved almost exclusively to this architecture. For storage, SCSI
remains a predominant protocol and is carried on a Fibre Channel
topology between devices. Now, with Gigabit Ethernet maturing, it
has become the conduit between video server gateways and archive
systems.
Given the reemergence of IP in video server systems,
it may be time to visit the viability of other principles in networking
and storage management for future expansion of the video server
system enterprise-wide. To do this, we’ll look at one of the
newer entries that is allowing the IT world to meet the needs of
expanded storage and access on a scalable and wider area network
topology.
Fibre Channel, originally developed for the efficient
connection of storage to servers, has evolved into a multitude of
new and expanding performance values. Fibre Channel’s major
strengths include its low latency, high-performance capability of
lengthy connection distances in a campus environment, with wide
availability from vendors specializing in storage and interconnection
products.
In addition to its ability to connect over longer
distances than its SCSI counterpart, Fibre Channel further addresses
the needs of scalability and availability.
Fibre Channel Arbitrated Loop is now widely understood.
This stable system provides for redundancy as well as a switched
fabric that allows for systems designed with both alternate path
routing and zoning of the data storage architecture.
STORAGE MANAGEMENT
Newer 1 Gb products provide for full-duplex throughput
of 200 MBps; and the Gb products on the market double that. Already,
a 10 Gbps Fibre Channel system – on the horizon in both product
and implementation – is showing promise.
The newest classes for Fibre Channel are in its storage
management domain. The storage area networking – SAN –
management emergence has been changing the architecture, and the
topology, of the storage model.
The variety of available products, choices of solutions,
and alternatives in technology will do much toward bringing the
SAN model more into the mainstream.
One drawback is the distance limitation inherent in
an FC SAN – holding storage applications within the confines
of the data facilities or – at the extreme – campus-wide.
This situation is unlikely to change over the next several years,
curbing widespread deployment in FC SAN beyond a very local area.
One solution to the confinement of SAN-type storage
over Fibre Channel lies in IP storage. Using conventional IP methods
and IP storage applications, the FC SAN can be extended over greater
distances. In mid-2001, the Internet Engineering Task Force (IETF)
worked on a standard that uses IP storage as an alternative to Fibre
Channel when implementing expanded SANs. The specification, called
iSCSI, is a protocol that defines how SCSI and Ethernet are harmonized
so that SCSI transfers can be made over TCP/IP networks.
Block level storage transfers are now available anywhere
on the IP network and – once again – the network bandwidth
now sets the throttle for data movement between devices.
iSCSI gives access to a centralized, high-performance
SAN that is familiar in principle and operations to many. As a compliment,
iSCSI for the IT environment expands the capabilities of the Fibre
Channel SAN over much broader areas – whether on the public
or the private network.
Fortunately, applications being deployed in the 1
Gbps domain work well over both 1 Gb Fibre Channel and 1 Gb Ethernet.
Gigabit Ethernet development continues and is now
well defined through 100 gigabits per second. With 1 Gb Ethernet
growing in popularity, 10 Gb Ethernet development also continues,
and broad deployment is expected in the next two to three years.
For small organizations – where the high cost
of building large-scale, robust Fibre Channel networks makes little
economic sense – building SANs around the iSCSI specification
will become attractive.
CENTRALIZED OPERATIONS
As the broadcast community continues its experiment
in centralized operations, more media will need to be stored in
a more distributed method. The movement of that information between
sites and organizations continues to be one of the more costly and
confusing issues surrounding centralized operations.
Moving toward less complex network-based media management
systems will be a driving factor in the implementation of systems
that more closely follow the IT domain or environment. It will be
interesting to see how standards, such as iSCSI, will change the
complexion of the video server architecture as faster IP-based networks
are deployed in myriad business and media applications.
Karl Paulsen is vice president of engineering at
AZCAR (http://www.STIDigital.com.)
He is the author of the book "Video and Media Servers: Technology
and Applications"– now in its second edition (published
by Focal Press). Contact him via email at kpaulsen@STIDigital.com.
|
