8   Sizing Up Servers

While you are destroying your mind watching the worthless, brain-rotting drivel on TV, we on the Internet are exchanging, freely and openly, the most uninhibited, intimate and, yes, shocking details about our “CONFIG.SYS” settings.

—Dave Barry

One of the most commonly used devices in digital video production and delivery is the video server. Almost all content ends up on a server during part of its lifecycle, whether for production, delivery, archiving or playout. Each of these applications has its own set of requirements and group of manufacturers offering specialized products. Because servers can be some of the more expensive items to purchase in an IP video delivery service, it is important to understand the demands of each type of application.

In this chapter, we will begin with a brief description of the major types of servers used in video applications. Then, we will examine a few of the categories in detail as they pertain to IPTV and Internet Video. We will conclude with a table that compares the key performance parameters for several types of servers.

Video Servers

Video servers perform two main functions: storage and delivery. Storage is the physical act of keeping files of digital video content (usually on hard disk) for processing or playout. Delivery is the act of transmitting video content over a network to viewers or other devices that need the content. Depending on the application, servers may be optimized for one task or the other or may need to strike a balance between them.

Video servers are often made up of a number of physically separate hard disk drives and processors. This is done both for greater reliability and for better performance. Reliability is increased through the use of RAID (Redundant Array of Inexpensive Disks) technologies, which store extra data for each file. This extra data can be used with a simple algorithm to replace any data lost due to a failure or replacement of one of the disk drives. Multiple disk drives are used to increase the storage capacity of the total system beyond what is available on a single disk and also to increase the speed at which files can be written or read from the disk array. Similarly, multiple processors are used to enable the system to continue operating even if one processor fails and to provide greater computing power than would be available from a single processor.

Here are quick descriptions of some of the different applications where video servers are commonly used.

•   Ingest servers are used to collect content from a variety of sources and make it available for use in a variety of applications. Video content can come directly from a studio camera or satellite feed, from a videotape that has just been removed from a camera or from archival storage, from another storage device such as a hard disk inside a camera or a remote server or from essentially any other source that can produce a video signal. Once the video has been ingested, it can then be handed off to a variety of other devices for further processing and storage.

    One of the most important roles of the ingest server is the proper tagging and description of each of the ingested video files. This information, called meta-data, which can be produced automatically but usually requires human intervention, is crucial to the later processing and manipulation of the video content. If, for example, a mistake is made in the date of video file, then an editor looking for the latest version of a shot may not be able to find it. High quality in this data can be aided by the software that operates inside the ingest server to support rules for capturing this data and processes that require proofreading of the data by a second person once it has been entered.

•   File servers are used in the video production process to handle content that is being manipulated into its final form. For example, a file server may be used to temporarily store a video clip from a color correction workstation before it is moved to another workstation that will be used to overlay graphics. The file server may also be used to store content or other data that is used repeatedly in the production process, such as theme music for a recurring program or common graphic elements.

•   Production or playout servers are used to take finished video content that is ready to air and play it out in a continuous, highly reliable stream. With these servers, reliability is key, because any failures can cause a broadcaster to go off the air. Various technologies can be used to provide redundancy and fail-safe operation; these features are commonly found on this type of server.

•   Archive servers are designed to store massive amounts of content. This can be from all types of sources, such as live feeds, news clips, purchased programming, etc. Archive servers typically emphasize large amounts of storage at a low cost, with speed of access being a secondary consideration. Archive servers can also be used to keep video records of programming as it has actually been broadcast in order to comply with local government regulations and to answer queries from advertisers.

•   Video on Demand servers are designed to store content that viewers can order for viewing. These servers are typically designed to generate as many simultaneous streams as possible, often multiple copies of the same content. High bandwidth network connections are almost always used, whether the connection is to a private IPTV network or to the Internet.

•   Advertising servers take advertising spots and play them back live inside video feeds. Although they typically don't need a massive amount of storage, they do need to interface to multiple simultaneous video channels and carefully synchronize the content playout to fit into the allotted advertising window. These servers need to be able to accept video content in a variety of different formats from multiple sources. In addition, these servers need to provide flexible scheduling tools that can be easily reconfigured to comply with rapidly changing advertising campaigns and keep good records of the ads that have actually run to support advertiser billing.

•   Live streaming servers take live video streams and create multiple copies for transmission on the network. Although they need practically no storage, they need to have a large amount of processing capacity to create IP packets that are individually addressed to each recipient of the stream. Live streaming servers also need high bandwidth network connections to transmit all of the streams that they generate out into the IP network.

The last three types of servers listed above are often used in IPTV and Internet Video applications, so we will go into a bit more detail regarding these devices in the following sections.

Video on Demand Servers

Video on Demand (VOD) is a common form of delivery for both IPTV and Internet Video networks. By allowing users to select content from a library at any time, this technology can be a powerful draw for attracting viewers to a service provider. It can also be a competitive weapon against services that rely on broadcast distribution, such as satellite and digital terrestrial networks. Most IPTV and Internet Video services providers, as well as many CATV systems, offer VOD.

VOD servers must perform four main functions:

•   Video content storage, which is essentially the same function as any other video server. However, the server must be capable of transmitting multiple, asynchronous copies of a single piece of content (more on this later).

•   Network interface, which again is similar to other video servers, with the exception that a very large number of simultaneous streams may need to be supported.

•   User interaction support, which enables a viewer to pause, rewind and fast-forward video content. This can require some sophisticated software to manage all of the viewers and to interface to the systems that process user commands.

•   Catalog and ordering support, which provides support for the systems that are used to display the list of available content, as well as the transactions necessary to capture payment from the viewers.

Content on a VOD server is essentially always stored in a compressed format that is ready to deliver to the viewer. This simplifies the delivery process by eliminating the need to process the video before it is delivered. Since many IPTV systems have a narrow range of allowed video signal rates and normally support only one (or at most two) compression formats, all of the content stored on the VOD server must be stored in the same format.

Accordingly, all incoming content must pass through a video compression device before it is placed on the server. In some cases, the content owner does the compression, and compressed files are simply copied directly into the server. In other cases, content may arrive in an uncompressed format and must be compressed before it can be placed on the server. This compression can be done in real-time as the content is streamed or it can be done off-line on a file basis.

In still other cases, content may arrive compressed using a different bit rate or type of compression. If an incompatible format is delivered, transcoding is used to convert it to a compatible format. If the bit rate of the content needs to be changed, transrating is used to convert the content. Note that transrating is normally only done to reduce the bit rate of video content.

When purchasing a VOD server, it is important to match the capabilities of the server to the task that needs to be performed. The amount of storage can be large or small, and the number of streams supported can be large or small. These are not correlated; it is perfectly sensible to have a server with lots of storage yet little streaming capacity, if it is being used to hold video content that is rarely viewed. Conversely, it is also sensible to have a server with relatively little storage (say, 50 to 100 hours of video content) but very high stream capacity, if it is being used to serve first-run Hollywood movies to many viewers simultaneously.

FIGURE 8.1   Centralized versus Distributed Servers

IPTV service providers have two main philosophies of network server distribution, as shown in Figure 8.1. The first is centralized, where large, high-capacity servers are installed in central locations and the streams for each viewer are delivered over high-speed communications links to each local service provider facility. The second is decentralized, where smaller servers are located at each facility and provide streams only to local viewers. A central library server provides content to the distributed servers whenever necessary. On one hand, the decentralized concept makes sense because it helps reduce the amount of bandwidth needed between locations. On the other hand, the centralized concept is appealing because it reduces the number of servers that must be installed. It also reduces the costs of transporting, storing and managing redundant sets of content in multiple locations. In reality, both centralized and decentralized systems are deployed depending on system architecture, capabilities and user viewing habits that affect VOD traffic patterns.

Service providers need video servers capable of delivering video streams to hundreds or thousands of simultaneous viewers. For this application class, specially designed servers are required. These units typically have a large number of disk drives and use multiple processors in parallel to format streams and deliver the content. The capacity of these systems is staggering; in order to supply 1,000 simultaneous users each with a 2.5 Mbps stream, the server needs to be able to pump out 2.5 Gbps of data. Since no single disk drive or processor in a typical server is capable of this amount of data, servers use load-sharing among the devices. This means that each piece of content is spread out across multiple disk drives and that a high-speed backplane interconnects the different drives to the different processors.

Calculating Server Storage Capacity

In order to properly calculate the amount of storage needed for a VOD server, two things must be known: 1) the number of hours of content to be stored and 2) the nominal bit rate of the video signal. With this information, calculating storage capacity is fairly straightforward.

Let's look at an example. Consider a one-hour video signal (with accompanying audio, of course) that runs at a bit rate of 2.5 Mbps. Recognizing that there are 8 bits in a byte and 3,600 seconds in an hour, you easily calculate that the total file will be 1.125 billion bytes, or about 1.05 GB. Note that this value is approximate, since the exact format of the file on a hard disk will be different. In addition, a small amount of metadata will be added to the file to provide description of the video and make it easier to transmit in multiple copies.

Here are a few more examples of required video server sizes for various amounts of content:

•   200 hours of SD content at 2.5 Mbps = 210 GB

•   500 hours of SD content at 4 Mbps = 900 GB

•   10,000 hours of SD content at 2 Mbps = 9 TB

•   300 hours of MPEG–2 HD content at 14 Mbps = almost 2 TB

•   500 hours of H.264 HD content at 6 Mbps = 1.35 TB

It is also interesting to note that some content owners place limits on how much compression can be applied to their video streams. Sometimes there are even contractual terms regarding the type of compression algorithm to be used. These limits are put in place to help ensure that the end viewer receives a high quality image. This can be very important to large production companies who have a public image to maintain and who stand to lose credibility or viewers if their products are over-compressed. One example of such a company might be a broadcaster who holds the rights to a large number of sporting events. If local IPTV providers use excessive amounts of compression, then not only will that group of local viewers get an inferior video feed, but there could also be a negative impact on the broadcaster's brand image in other aspects of their business.

Advertising Servers

An advertising server can be a key revenue producer for IPTV systems and may also have a role to play in live streaming Internet Video applications. The server's job is to insert advertisements into video streams at specially indicated times called avails. The result is a video stream delivered to a viewer with specialized advertising inserted.

Let's look at an example of how this technology works. A national broadcaster such as CNN designs their programming to accommodate advertisement inserts throughout the day. Many of these times will be sold to national advertisers and broadcast by CNN to every viewer in a country. Other times will be made available for local providers to sell to local advertisers. During these times, CNN will include audio tones or special digital codes that indicate that these times are available for local ads to be inserted. The ad server will recognize the indicator and replace the feed from the network with a video file stored on the local server. Any viewers watching CNN through the local provider will see the local ad in place of the ad broadcast by CNN. Because the timing of each avail is under CNN's control, the network can make sure that local ads are not inserted in place of high revenue national ads, but rather in place of ads that may not bring direct revenue to the network, such as advertisements for upcoming programs.

In the case of Internet Video, ads are commonly delivered in two ways. One way is as a graphic or video clip on the Web portal, where viewers navigate to select the clip they want to view or download. The second way is as a video spot advertisement delivered to the viewer immediately before or sometime during the requested content.

From a business standpoint, local advertising can be a big source of revenue to any video delivery system operator. Over-the-air, CATV and satellite broadcasters all utilize this technology, and IPTV and Internet Video operators can earn revenue as well. This revenue can be used to help offset the costs of programming and delivery systems, such as IPTV networks or Internet Video servers. Both local and national advertisers will use local advertising for certain purposes. For example, it makes no sense for a local automobile dealer to advertise on a national basis. National advertisers may also want to deliver advertisements selectively to local audiences, such a beverage company that may have an advertising tie-in with a local sports team.

Live Streaming Servers

Live streaming servers are used to support broadcasts over the Internet. They are necessary because each video stream delivered must be made up of packets specifically addressed to each individual viewer's device–there is no mechanism on the Internet to make copies of a video stream and deliver it to multiple users (i.e., multicasting). Another way to describe a live streaming server would be as a unicast replication server, because their principal job is to take in one unicast stream, make multiple copies and then send them on towards multiple viewers.

Unicasting is the standard mode for sending packets over the Internet. In this mode, each packet has a single source address and a single destination address. If a source wants to send packets to multiple destinations, it must create a unique packet for each destination. This requires processing power, because each packet needs to have a correctly formatted header, with a destination IP address, a correct set of flags and a properly calculated header checksum. Once the packet is created, it flows essentially intact directly from the source to the destination over the Internet.

Live streaming servers need very little storage, because the content is moving through in real time. Instead, these servers need a lot of processing power, because they need to receive incoming streams, make copies for each viewer and create properly formatted IP packets in a continuous stream for each viewer with little or no delay. In addition, the servers must be capable of processing transactions to add and drop viewers as people tune in to watch the video or tune out when they have seen enough or want to switch to other content. These servers may also need to capture data as needed to produce invoices for paid content, although that task is normally the responsibility of the Web portal that authorized the user to view the video.

In contrast to most other types of servers, live streaming servers don't have to be purchased by each company that wants to use them. Instead, service bureaus will (for a fee) provide processing power and Internet bandwidth when a company wants to host a live event. These bureaus, often called Content Delivery Networks (CDNs), will also host normal Web site content for delivery to Web surfers located around the Internet.

Encryption and Rights Management

Purchasing and installing a major server system can be a challenge. But, getting the rights to enough content to fill the server can be a much more daunting task. Owners of the content will often refuse to permit their programming to be placed on a server until they are satisfied with the security arrangements. Securing these rights often involves direct negotiations with the content owners and may depend on the certification of the DRM system.

There are a number of vendors of DRM systems who have taken the necessary step of proving the security of their systems to the satisfaction of major content owners, such as Hollywood movie studios. At a minimum, a DRM system must ensure that the content is unusable (for viewing or copying) unless the viewer has been provided with the proper key. There are a number of mechanisms for controlling and distributing these keys, which were discussed in Chapter 7.

DRM is not only important in a VOD delivery network, but also for the content storage itself, directly within the VOD server. This is to prevent unauthorized uses of the content, which could occur from an outside intruder gaining access to the server or from an inside user misappropriating the content. Content owners will typically insist on protection of their property both in storage and during delivery.

VOD vendors have taken a number of steps to protect stored content within their systems. In addition to standard encryption techniques used in DRM, some vendors have developed a proprietary file system that is separated from the normal server operating system. This can help prevent hackers and viruses from reaching the stored content. A second security technique involves breaking the content up into small files and distributing those files to physically separate hard drives. In the event that one of the drives is stolen or compromised, the content is useless, because it is only a small portion of the overall file. This system also provides extra reliability, because files can be stored along with error correction data so all files can be properly reconstituted even after a drive fails.

Reality Check

In this chapter's Reality Check, we will discuss three novel server implementations. In the first two examples, we will discuss ways to use servers to increase revenue and in the third, a way to change the physical location of the stored video content.

Selling Space on a VOD Server to Advertisers

On most VOD servers, there is a significant amount of space allowed for expansion. While this space may eventually be used up, for a good part of many system lifecycles, the space is simply empty. Some clever system operators have figured out a way to leverage this asset: selling space to advertisers.

In this situation, the advertisements are not the normal 30- or 60-second spot ads. Instead, they are long-form ads designed to appeal to the relatively small proportion of viewers who might want to get more information about a specific product or service. For example, a manufacturer of an innovative flooring product might want to sponsor an instructional video that shows consumers how easy it is to install and maintain their product. Or, a luxury automobile manufacturer might want to host a program that shows a classic car rally featuring their products. Or, a golf equipment manufacturer might want to sponsor a golf training video. Or, the visitors’ bureau for a tropical island might want to host a tour of their natural features. There are many possibilities.

To make this successful for both system operators and advertisers, a few conditions must be met. First, there must be a way for viewers to find out about the content and navigate to it. This will certainly involve the use of listings in the interactive program guide, but may also involve splash screens or inserts in more popular pages to inform viewers that the content exists. Second, the system operator may want to exercise a minimal amount of editorial control, to help ensure that the sponsored content doesn't end up being larded with hard-sell infomercials that have little value for viewers. The system operator may also want to gather some viewing statistics to see which types of content are popular and to provide feedback to the advertisers on the effectiveness of their offerings.

Ads Attached to VOD Content

As discussed above, not all VOD content needs to be paid for by viewers on a pertransaction basis. As we discussed in Chapter 3, many different models can be used to pay for on-demand content. Deciding how to implement an advertising-supported VOD system can be quite interesting.

One of the most basic decisions that must be made is to decide when the advertisements will appear. Many viewers have become accustomed to pre-roll advertisements, where a few short spot ads are played before the desired content begins to play. This is, of course, common practice in movie theaters (previews of coming events, reminders to not smoke and advertisements of the snack bar in the lobby are common themes). Pre-roll advertising is also common on Web sites and is part of many purchased content items such as DVDs and VHS tapes. The secret to not upsetting viewers is to ensure that the ads are brief and few in number.

A more controversial form of advertising consists of advertisements actually inserted into the content itself. This technique certainly does not appeal to some viewers. However, if the service provider makes it clear that advertising is required in order to pay for the content, then viewers are more likely to understand.

One very controversial aspect of advertising and VOD content is whether commercial zapping should be allowed. Commercial zapping occurs when a viewer decides to fast-forward past a commercial. Most live video recording devices (such as personal video recorders) allow users to fast-forward through advertisements.

The two sides of the controversy regarding VOD services can be summarized as follows:

•   If zapping is not allowed, then users who are not willing to see advertisements will become less likely to watch the content. This in turn could translate into fewer overall viewers, meaning that revenues that depend on viewers (such as subscription fees) will drop.

•   If zapping is allowed, then advertisers will be less likely to pay for advertising, since there is a lower probability of their ads being viewed. Service providers may then find it necessary to charge more for VOD content.

Of course, it is not necessary that this decision be made on an all or nothing basis– service providers are free to have different amount of advertising for different types of content. They can also experiment both with advertisements that allow zapping and ads that don't, at the risk of truly confusing viewers.

Push VOD as an Alternative to Centralized Servers

Push VOD uses hard disk storage located inside the viewer's STB to store content locally that can be viewed on demand. Push VOD is a relatively new concept being used to provide VOD over networks that don't have interactive capabilities. This is certainly the case with satellite networks, which simply don't have the bandwidth to create a separate video signal stream for each user.

In an IPTV system, push VOD might be useful for a few reasons. First, by storing video files locally in each user's STB, the burden on the network could be reduced when a viewer is watching a VOD program, and the load on centralized VOD servers is lightened. Second, locally stored content could be used to provide entertainment or troubleshooting information to users in the event that their network connection failed. Third, local storage could be used to provide highly interactive programming or entertainment (such as games) that would be difficult or impossible to provide from a centralized server.

Of course, there are some factors that must be considered before using push VOD. First, a very strong DRM technology will be required, since push VOD content is literally sitting in a hard drive in the viewer's home. Second, a fairly sophisticated control system will be needed to manage which content gets delivered to each STB and to collect the payments from viewers that chose to watch the content. Vendors are appearing that offer to manage both of these issues for service providers, and as hard drive capacities increase, more content can be stored.

One interesting concept that can benefit both viewers and service providers is installing a partitioned hard drive in an STB. In one partition, the service provider can push a dozen or two popular movie titles being featured for VOD. The other partition can be used to give the viewer a PVR capability for their favorite broadcast shows. This combination gives system operators two ways to pay for the extra expense of purchasing and maintaining hard drives in STBs–by selling push VOD content and by increasing the rental fees to viewers for STBs that include PVR capability. STB manufacturers are starting to respond to this market by placing large hard disk drives into STBs and supporting partitioning in the STB operating software.

Summary

In this chapter we have discussed a variety of different server types and examined in detail three types often used for IPTV and Internet Video systems. VOD servers can be large or small, centralized or distributed, but they are always rated on the number of simultaneous streams they can support. Advertising servers are typically not large or hugely powerful, but they need to be able to monitor multiple live network feeds, reliably insert ads and keep good records. Live streaming servers need almost no storage but are rated like VOD servers on their total throughput in terms of number of simultaneous streams. Table 8.1 summarizes some of the similarities and differences between the various types of servers.

SERVER TYPE	CAPACITY	SPEED	COST	KEY ATTRIBUTE
VOD	Varies	High stream capacity	Low to moderate	Bandwidth–number of simultaneous streams
Archive	As large as possible	Not important	Lowest cost per terabyte	Large capacity at low cost
Playout	Low	Low	High	Reliability/redundancy essential
Advertising	Low	Able to handle multiple channels simultaneously	Medium	Easy to operate software, excellent record keeping
Live Streaming	Very Low	High stream capacity	Medium	Bandwidth—number of simultaneous streams
Ingest	Low	Low	Medium	Flexibility for video inputs, good software for metadata workflow

TABLE 8.1   Key Attributes of Different Types of Servers

 

1. Seagate press release http://www.seagate.com/www/en-us/about/news_room/press_releases/