Delivery of Digital Files--Network Connection Environments
From MODwiki
Contents |
[edit] Overview
University of Utah Campus VOD Network Architecture Considerations
One of the biggest bottlenecks facing the University of Utah campus network will be the connectivity from the distribution nodes to access layer switches. Almost half of the buildings on campus are still connected with 100mb networks and some of those buildings are still running older network technologies that will prevent them from benefiting from a faster 1gig link. (token ring, hub non switched networks, etc.) Most of the high bandwidth buildings are already 1gig and have the inside infrastructure to support that all the way to the classroom and office. But there are some exceptions, for example, the Eccles Broadcast Center is not gig connected.
From our current campus Core nodes (one in Research Park and one in EBC) to our EBC distribution node, we have 4 gig's full duplex. Plans for 20 gig full duplex are in process awaiting several new power supplies and PDU's to be installed in the node rooms. The Campus data centers are today all 4 gig's connected, but there are no plans today on getting them to the 20 gig interface.
Any VOD Network infrastructure plans should make a case to be directly connect to the core networking equipment OR define that the data center networks be at the 20gig level before deployment to prevent network core/dist architecture from being a possible bottleneck. Also some protocol research should be done to define which and what protocols will be used to distribute content across campus.
For example on today's network connected to a high bandwidth building we could support about several 4meg streams without network latency. Into the building networks is where this network would start to have problems. Most campus building networks are running near the 40% Ethernet capacity for internal traffic.
Ken Couch, Director of Marketing for Broadband Access Nortel says, “When designing the optimal Campus VOD network infrastructure, there are many aspects that need to be considered. Some of the key decision points include:
- How much storage will be needed based on amount, length, and types of content?
- What type of storage and streaming technology is best? Hard drive vs. DRAM?
- Is it better to deploy using a centralized or distributed Campus VOD architecture?
- How much bandwidth does my network need at various points in the network?
The general starting point is to build a model that forecasts the expected take rate and usage patterns of a Campus VOD service, including a determination of the peak Campus VOD usage rates. This information is then used to further build out the model to determine how much storage is needed for library content (asset storage), how often must the content be refreshed on a regular basis (asset ingest), and the expected peak bandwidth requirements to reliably deliver the service.
It is important to note that Campus VOD is unicast in nature, which means there is a dedicated content stream for each end-user session. Multicast has limited Campus VOD uses with the possible exception of multicasting content in a distributed architecture to cache servers. The unicast requirement places significant bandwidth demands on all parts of the supporting infrastructure.
Finally, it is important to determine the type of Campus VOD technology that is best optimized for the long term Campus VOD service. If Campus VOD continues to migrate from a traditional Movie-on-Demand service towards Television-on-Demand (ToD), or for that matter an Everything-on-Demand model, the implications are that content ingest rates will skyrocket. Video content will need to be refreshed and updated continuously to the point where it reaches a near real time ingest rate and then be broadcast to thousands of users simultaneously. From a subscriber perspective, this would be like having a PVR service, without the need for a PVR device in the home.
The implications of a dramatic increase in asset ingest rates is that there will have to be a corresponding response in Campus VOD server technology and network designs that are optimized to handle this shift. Today most Campus VOD platforms use some type of hard-drive technology to store and stream content. Hard drive technologies are low cost, and they do a great job of streaming content out to users, but the close coupling of the storage and streaming functions presents technical challenges as the library grows. The newer DRAM based platforms have the benefits of higher reliability, lower power, higher density, and the ability to handle the rigorous ingest / distribution demands of a Television-on-Demand environment. Their drawback is higher cost, particularly for low scale deployments.
It may be that the optimal Campus VOD architecture will be some combination of hard drives for whole-library storage and using DRAM to stream the high-usage content. DRAM has the potential advantage of being able to supporting many concurrent streams due to its ability to pump out content at very high bit rates. The optimal network configuration / design is obviously dependent on the individual requirements of each service provider and the scale of the network.
[edit] Guidelines
Distribution Network
Video on Demand (Campus VOD) requires predictability and continuity of traffic flow to ensure real-time flow of information. MPEG and MPEG-2 require an effective BW of 1.5 - 4 Mbits/sec. Multiplying this "media stream" BW requirement by the number of clients will give a rough estimate of the effective distribution network's bandwidth. The Common Imagery Ground/Surface System (CIGSS) 1 Handbook suggests the following steps to size and specify the LAN technology use for Image dissemination systems:
- Approximate the system usage profile by estimating the amounts of image, video, and text handling that will be required.
- Convert the amount of images, video, and text to be processed into average effective data rates. Raw data transferred directly to an archive (our video server) and near real-time processed imagery should be estimated separately. The bandwidth requirements can be combined later if needed.
- Adjust calculated rate for growth. The growth factor should be at least 50%.
- Add a fraction (about .3 to .4) of the peak capacity to the growth adjusted rate for
interprocessor communications.
Updating heritage networks to this new BW requirement can incur substantial costs. The cost of implementing a hi-speed network varies depending on the network architecture.
There are several transport protocols that can be implemented for audio-video applications; TCP, UDP, SONET, TCP/IP Resource Reservation Protocol (RCVP). Do to the effective data rate necessary to support Campus VOD, protocols that minimize client/server interaction are preferable, except in cases where an over-abundance of network bandwidth exists. In Ethernet nets supporting mostly non-Campus VOD applications, retransmission of lost packets or corrupt packets will not be possible. For example, if packets are lost they uses pixel tiles from a previous frame. In a typical Campus VOD system, without error correction, QOS is directly proportional to network/LAN BER (Bit Error Rate). Campus VOD systems which provide error correction as part of network protocol have to be designed to allow for the latency created by their error correcting protocols. (DSS currently implements interleaving, Reed Soloman and viterbi decoding) QOS trade-offs can be quantified and analyzed.
[edit] Resources
- Akamai Whitepaper-- "Best Practices in Digital Media Delivery"
- http://www.akamai.com/html/perspectives/whitepapers_content.html
- Akamai Whitepaper-- "Akamai Streaming--When Performance Matters"
- http://www.akamai.com/html/perspectives/whitepapers_content.html
- VitalStream-- Showcase of Media On Demand Clients
- http://www.vitalstream.com/showcase/index.html

