Verizon Wireless deployed video optimization technology in parts of its 3G and 4G mobile broadband network. This network management technology is designed to transmit data more efficiently, ease capacity burdens on the network, primarily from video files, and improve the user experience with faster downloads and decreased Internet latency.
Given the increasing web traffic for downloading video files, video optimization in particular benefits both the user as well as the network by facilitating sustainable online video browsing. By downloading only the necessary amount of data, video optimization enhances the video download experience as well makes room for other users to enjoy higher browsing speeds. Although much effort is made to avoid changing the file during optimization, the process may minimally impact the appearance of the file as displayed on a device, though changes to the file are not likely to be noticeable.
The optimization techniques are applied to all content files coming from the Internet Port 80 that use the most common compression formats. The form and extent of optimization depends on the compression format of the content file, but does not depend on the content of the file, the originating web site, or the user's device. No distinction in the application of these techniques is made based on the source website or originator of the content. The system optimizes files based strictly on the type of file and the relevant file formats (recognizing that some file types are not modified). Accordingly, all content, including Verizon Wireless branded content, of the same type will be subject to the same process.
Why Optimization? Delivering content files requested by an end user over the Internet always imposes some burden on the delivery network in terms of size of the file as well as the distance the file components must travel between the source and end user. These factors also directly affect the user experience in downloading the file.
When the network uses techniques to "optimize" or streamline content files the burden on the network can be lessened and both the speed and efficiency of delivery to the end user can be improved. For example, the size of the file can be compressed by removing pieces of information that are not usable by the end user's mobile device, or that are not noticeable to the user. Caching the file for subsequent requests can also reduce the time needed for delivery to end users. Such network management techniques improve the user experience without noticeable impact on the content itself.
Guiding Principles of Congestion Triggered Video Optimization. The goal of video optimization is to only optimize video files when necessary and to make intelligent tradeoffs that maximize the total quality of video to all users. The objective is to be proportionally fair as measured (or estimated) in total video quality viewed on the customers device. For example, video stalling is objectionable to the end user. Therefore, a standard resolution video that does not stall is perceived to have a higher total video quality score that that of a high definition video that does stall. An intelligent network decision that reduces the resolution of a video to prevent stalling is deemed to be a good tradeoff for customer satisfaction.
How Optimization Works. All HTTP (Port 80, i.e., World Wide Web) traffic is directed to the optimization process. The first step is to determine if the serving cell site is highly loaded, indicating that any new video flow is likely to be degraded and/or stall. If there is no potential congestion, then nothing is done and the video proceeds normally. If there is a potential for stalling, then the video optimization techniques are invoked. The direction of traffic to the optimization process is established when the user starts an HTTP data session before any requests for content from a specific web site have been made. Accordingly, content files are never selected for optimization based on the nature of the web content itself or the source or provider of the web content file. All web content files delivered over Port 80, regardless of source, are directed to the optimization process (assuming the cell site congestion criteria are met). The system captures all Verizon Wireless branded web content delivered from its web servers, and treats it the same way as content from non-Verizon Wireless sites.
Content files made available on the World Wide Web come in a variety of types (web pages, text, images and videos) and formats. The process uses several optimization techniques that depend on the specific type of content file. Specifically, text files are compressed without any loss of information ("lossless") and cached for subsequent end user requests. Image files (PNG, JPEG, and GIF formats, for example) are streamlined to remove colors or other data bits that would not be visible to the human eye, or to end users on a mobile device with limited display resolutions, thereby decreasing the size of the file, and are also cached. The output image file reflects "lossy" optimization because some data bits from the original file are lost in the optimization process.
Video Optimization. Video files represent a substantial and growing segment of web traffic, and also come in a variety of formats. Optimization only captures recorded video files and does not affect live streaming video, e.g., a video conference call. Several optimization techniques are applied to video files: transcoding, caching, transrating, buffer, and ABR tuning. All are agnostic as to the source or content of the video.
Transcoding. When preparing a video file for posting on a web site, the video originator must select a codec (compression/decompression format) for the file. All codecs are "lossy" to some degree in the compression process in that they reduce the quality of the original video. But, some codecs are more efficient than others. The Optimization transcodes video files from their source codecs to a more efficient codec, H.264. If the requesting device cannot decode an H.264 file, the file is delivered in the input codec. Also, if the input file codec is H.264, there will be little or no effect on the file from the processes described below.
The goal of this optimization process is to reduce the content file size while maintaining very similar video quality. Re-quantization levels, that is, the size of the output file, are defined by the output video bit rate settings (based on a percentage difference from the original). The loss of information from the input file may result in reduced color accuracy and sharpness of the output video. These effects are offset with optimized de-blocking and smoothing algorithms to retain good perceptual visual quality (as measured by objective video quality tools discussed below). In addition, videos are sent with variable bit rate (VBR), which provides more consistent quality at the same bit rate.
Optimization processes can range in how aggressively they pursue content file savings. Verizon Wireless is using the Video Quality Measurement (VQM) tool to set the amount of reduction in a video file size. VQM is a standardized method of objectively measuring video quality that closely predicts the subjective quality ratings that would be obtained from a panel of human viewers. Although the tool is free, the technology is covered by four U.S. patents owned by NTIA/ITS. The compression settings utilized equate to a .4-.6 score on the VQM scale, which is considered an "unnoticeable" change.
Transrating. This is similar to transcoding; however the source and destination codecs are kept in the same codec family. For example, the H.264 codec can have a higher input rate and be transrated to an H.264 codec with a lower delivery rate. The VQM treatment is similar to transcoding. Verizon Wireless selects the minimal amount of transrating to keep the overall video quality as high as possible while maintain the objective to not let the video stall. As H264 has become a prominent and commonly adopted industry wide codec, more instances of transrating exist over transcoding.
ABR Tuning. The newest high technology codecs automatically and continuously adjusts the video bit rate to match the estimated channel conditions. Adaptive Bit Rate (ABR) types of codecs performance can be improved by tuning the actual network conditions to the various codec rates supported for each video.
Caching. When a video file is detected from the Internet stream, the system decodes the first few frames (8 KB) of the video. Based on those frames, the system attempts to locate the video file in its video cache, and, if the file is not in the cache, it copies the video file, catalogs, optimizes and places it into the video cache. (The system needs to look at the first few frames for the cataloging process because the same video may come to the network from different sources and would have different URLs and headers; so, the header information is insufficient to identify multiple copies of the same video.) The caching process is the same regardless of the source or content of the video.
When a requested video is not in cache initially, the input video file is sent on to the requesting device. When the system finds the video in its cache, then the flow from the Internet stops and the video is replaced with the file from the cache. The video cache will retain the video until the staleness filter flushes it from the cache. The video cache has a finite volume so it will regularly flush unused video.
Buffer Tuning. The fourth video optimization technique is used in delivery to end users. Whenever the video is requested, it is delivered on a "just in time" basis. That is, rather than the entire file being downloaded when requested, the video is downloaded on an as-needed basis. A sufficient amount of video would be delivered to fill the user's buffer to start viewing, and the remainder would be delivered as needed in time for the viewer to see it without interrupting the flow, calculating the video bit rate and the actual bandwidth available. This progressive download achieves significant network savings if the viewer chooses not to view the entire video, and it conserves data usage that would count toward the end users' data allowance, and may result in savings if the end user is on a pay-for-usage plan. As with caching, the buffer process is the same regardless of the source or content of the video.
These video optimization techniques generally reduce the time for a video to start and eliminate external network fluctuations that sometimes cause videos to stall. They also speed up the time for the video to pick up when jumping forward in the video. The cache responds to the video request much faster than a remote location. The end result is a much smoother video that starts faster.
Beginning in 2011, to optimize our network, we managed data connection speeds for a small subset of customers - those who are in the top 5% of data users and have 3G devices on unlimited data plans - and only in places and at times when the network was experiencing high demand. We discontinued this practice in June, 2015.