Video to the desktop.......................................
telecoms-mag.com
As desktops become more powerful and video-enabled with Intel's MMX processor for example, multimedia networking becomes the next logical step. Power users can clearly benefit from desktop video teleconferencing and remote collaboration. Getting a live video broadcast out to all desktops can be a powerful and convenient way to communicate important messages, and adding video content to your corporate intranet can be a great way to deliver training and information vital to work flow processes in an organization. Despite the benefits that video can deliver over networks, few applications have generated more confusion as far as implementation is concerned. This article focuses on some of the work that's been done on multimedia intranets and shows how low-cost switched routing can be a key component to supporting these kinds of networks.
In a multimedia intranet there are two keys elements that must be considered from a networking point of view. The first is bandwidth to the desktop and the second is routing across the enterprise, and they are closely related. When it comes to bandwidth to the desktop, Ethernet will continue to dominate for some time. This is generally fine for most networked multimedia, as will be shown below. However, the second key element, routing, must be rethought when it comes to networked multimedia.
This is because traffic flows in multimedia intranets are just the opposite of the "80-20 rule" which has governed routing over collapsed backbones for some time now. The 80-20 rule states that roughly 80 percent of your LAN traffic stays within a subnet, mostly sharing servers and printing facilities within a work group, while 20 percent leaves the subnet. The 80-20 rule went along with the axiom "bridge where you can, route where you must," since router ports were expensive, had performance issues when compared to switching, and were more complex to manage.
Clearly we have been shifting away from 80-20 for some time now, especially with the introduction of the Internet, but multimedia looks more like 5 percent-95 percent when it comes to intrasubnet vs. intersubnet traffic. This is because it is far more useful to video teleconference with someone out of your general area, i.e. across subnet boundaries. Also a video server used for training on demand is likely to be a centralized device, so again video will be streaming across subnet boundaries to reach interested users. Thus a multimedia intranet by definition will have to support much more traffic across subnet boundaries. Making routing more scaleable and compatible with streaming protocols such as video and audio is critical to successful networked multimedia. This is where switched routing over ATM comes into play.
Switched routing is based on the ATM Forum's MPOA (multiprotocol over ATM) standards activity and makes ATM subordinate to the routing function. ATM becomes the core of a highly scaleable, switched-oriented architecture that is ideal for enabling applications such as multimedia intranets
The idea for switched routing is to take high performance, standards-compliant ATM switches and make them the core of a routed infrastructure. Then run a protocol at the edges of this infrastructure that translate network layer addresses such as IP, to ATM addresses known as NSAP. Given that 80 percent of networked desktops are running Ethernet over 10Base-T wiring, an Ethernet to ATM edge device is key. This is where the address translation protocol operates, as well as Ethernet to ATM segmentation and reassembly. The protocol standard that encompasses switched routing is MPOA and went to a straw vote in the ATM Forum in February and was ratified at the ATM Forum in July of this year. MPOA takes the traditional packet router out of the forwarding path and allows for cut-through switching at the network layer to take its place.
The key to this paradigm shift is the collapse of what was formerly a pivotal scarcity and the rise of new forms of abundance. In the case of switched routing, high performance, affordable routing was the scarcity, and cut-through network layer forwarding with quality of service (QoS) options is now available in abundance. Making ATM subordinate to routing is essential for the 5-95 world of multimedia.
Back at the desktop, the most obvious concern is bandwidth, since few applications will generate more traffic than video and audio. Bandwidth directly relates to picture quality, but is only one of several key parameters related to overall performance. Latency (end-to-end delay), which is critical to desktop video teleconferencing, is another key parameter and one that is hard to control in routers. Video can be handled in hardware or software, with the trade-off being image quality, cost, and ease of implementation. Regardless of desired picture quality on the desktop, all video is digitally compressed by algorithms that are optimized for a given application. The result is that video to the desktop can live very comfortably within 10 Mbps. So it should be clear from a bandwidth point of view, that video to the desktop will run fine within an Ethernet segment.
What we have also found is that currently no operating system on a general purpose desktop machine can handle more than an aggregate of 4 to 6 Mbps of streaming video. Again, this is well within the range of a single Ethernet segment.
In practice, there are some other very good reasons why video streams to the desktop are under 2 Mbps for the majority of applications. In the case of video client/server applications, the largest single cost is video storage, so even where high quality is a requirement, the video bit rate is almost always 1.5 Mbps or less. This translates into roughly $500 per hour of storage for MPEG-1 files on a video server. At this rate there is a healthy incentive to use as much compression as possible, though the MPEG video quality at 1.5 Mbps is surprisingly good. Applications that can get away with lower quality will require lower bandwidth, again well within the reach of a dedicated Ethernet to the desktop.
Desktop videoconferencing has even more constraints on it. Since it's far more interesting to videoconference to a distant site, there will almost always be a high cost of WAN component. This generally limits videoconferences to 128 kbps. Again, this is a drop in the bucket for Ethernet. If I'm staying on the LAN, it is rare to have desktop videoconferencing running over 500 kbps. Even running an MPEG video clip from a server and a videoconference simultaneously on the same desktop (assuming the machine can be configured to handle it), results in under 2 Mbps of bandwidth. We demonstrate this all the time in the VIVID video lab and get the most incredulous looks when we tell folks we're running Ethernet to the desktop!
The issue then becomes how do we share an Ethernet segment among several video-enabled users and a router port? This is not an easy question to answer, since video generally runs under UDP and losing video packets as a result of collisions can impact quality. As a general rule, collisions on an Ethernet segment are minimized when utilization of the segment remains under 30 percent. This is probably a good number to use when analyzing potential video traffic loads in a segment for purposes of microsegmentation options. Now in the case of the switched router, the marginal cost of a 10Base-T port can be as low as $500, which approaches the cost of LAN switching. With the reduced cost of switched routing, it is not unreasonable for a power multimedia user to have a dedicated Ethernet segment directly into the router. Since subnets in a switched router are virtual and mapped to any number of ports, direct connections do not eat into address space as with a traditional router. Thus provisioning video-enabled users with 10-Mbps wire speed routing can be done by simply moving the desktop connection from the LAN hub or switch to a switched router port on an edge device in the same closet. Desktop NICs, protocol stack, and LAN wiring remain the same.
In the case of VIVID, we call this edge device a "ridge." Twelve Ethernet collision domains and one OC-3 155 Mbps uplink make up a single ridge. Each port delivers 10 Mbps of wire speed connectivity over a routed ATM backbone and a connection to any other point in the network. ATM switches offer the scaleability and QoS options necessary to handle large amounts of video. MPOA in turn allows network layer protocols such as IP to run over ATM. An example of a multimedia intranet is shown in Figure 6. Note that light duty users remain on their hubs and have no interaction with video-enabled users, though they all run over the same switched routing infrastructure.
The next step is for a client application to use a protocol such as RSVP (currently an IETF bandwidth reservation protocol draft) to signal the ridge. Noting the presence of RSVP in the stream can guarantee bandwidth and QoS in the ATM fabric without requiring end stations to be ATM attached. This does away with another myth about ATM, that it is required out to the desktop to make use of QoS options.
As more network planners and application providers experiment with multimedia-enabled intranets, the need for low-cost, scaleable network layer switching will become apparent. Routing is the key to the success of intranets, and making it cost effective and abundant to address the 5-95 world of multimedia is at the heart of switched routing.
Marc P. Pfeiffer is the director of VIVID Network Solutions at Newbridge Networks in Herndon, Va. Prior to joining Newbridge, he was president of Broadband Design Group, a Washington D.C.-based technical consulting firm that specialized in all aspects of video network design. Pfeiffer received his B.S.E. degree from the Moore School of Electrical Engineering at the University of Pennsylvania in 1982. He is a member of the IEEE Communications Society and the Society of Motion Picture and Television Engineers, and has spoken and written widely on the topic of multimedia networking. He is currently writing a book on the subject. For more information, visit the Web site, www.vivid.newbridge.com. |
