{"title":"Tile-based Caching Optimization for 360° Videos","authors":"Georgios Papaioannou, I. Koutsopoulos","doi":"10.1145/3323679.3326515","DOIUrl":null,"url":null,"abstract":"Panoramic, or 360° video streaming and playback are prime components of mixed-reality and panoramic movie-viewing experiences, they offer an immersive viewing experience, and they are becoming increasingly popular, albeit very resource-demanding. Caching of 360° video content at caches close to the user reduces content delivery delay and bandwidth consumption. In 360° video streaming, the recently proposed tiling approach allows the streaming of different parts (tiles) of the 360° content at different resolutions as opposed to monolithic single-stream transmission. We study the problem of optimal caching of 360° video streams. The difference from conventional video is that each tile may need to be cached simultaneously at multiple resolutions, since it may appear in different positions at different viewports, and the time proportions of these positions are dictated by users' viewing statistics. For example, a tile close to the center of a viewport must have high resolution, while the same tile close to the edge of another viewport can have low resolution. This leads us to a novel caching objective. Cached resolutions for each tile should be as close as possible to the required ones for each tile, according to tile viewing statistics. We study the cases of caching tile streams at different resolutions or in a layered encoding fashion. We seek to optimize an objective that combines (i) an error metric between requested and cached tile resolutions across viewports, and (ii) coverage of the tile set. For the case of multiple resolutions, we show that the problem of selecting the tile resolutions to cache so as to minimize the error metric above is equivalent to the K-Medoids problem. For layered encoding, the problem is to find the maximum-resolution layer to cache for each tile stream, and this is equivalent to a Multiple-Choice Knapsack problem. We present an implementation of the cache optimization scheme, we evaluate our model on a tiled 360° video distribution simulator and demonstrate a significant increase in cache hit ratio over conventional caching strategies.","PeriodicalId":205641,"journal":{"name":"Proceedings of the Twentieth ACM International Symposium on Mobile Ad Hoc Networking and Computing","volume":"88 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2019-07-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"33","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the Twentieth ACM International Symposium on Mobile Ad Hoc Networking and Computing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3323679.3326515","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 33
Abstract
Panoramic, or 360° video streaming and playback are prime components of mixed-reality and panoramic movie-viewing experiences, they offer an immersive viewing experience, and they are becoming increasingly popular, albeit very resource-demanding. Caching of 360° video content at caches close to the user reduces content delivery delay and bandwidth consumption. In 360° video streaming, the recently proposed tiling approach allows the streaming of different parts (tiles) of the 360° content at different resolutions as opposed to monolithic single-stream transmission. We study the problem of optimal caching of 360° video streams. The difference from conventional video is that each tile may need to be cached simultaneously at multiple resolutions, since it may appear in different positions at different viewports, and the time proportions of these positions are dictated by users' viewing statistics. For example, a tile close to the center of a viewport must have high resolution, while the same tile close to the edge of another viewport can have low resolution. This leads us to a novel caching objective. Cached resolutions for each tile should be as close as possible to the required ones for each tile, according to tile viewing statistics. We study the cases of caching tile streams at different resolutions or in a layered encoding fashion. We seek to optimize an objective that combines (i) an error metric between requested and cached tile resolutions across viewports, and (ii) coverage of the tile set. For the case of multiple resolutions, we show that the problem of selecting the tile resolutions to cache so as to minimize the error metric above is equivalent to the K-Medoids problem. For layered encoding, the problem is to find the maximum-resolution layer to cache for each tile stream, and this is equivalent to a Multiple-Choice Knapsack problem. We present an implementation of the cache optimization scheme, we evaluate our model on a tiled 360° video distribution simulator and demonstrate a significant increase in cache hit ratio over conventional caching strategies.