Yair explained that their Bytestream caching (background - here) solution promises "30-50% bandwidth expansion ratio with or without caching". The following examples show how it does it on top of file-caching:
- Monetized video: Many popular web sources (e.g. Rapidshare, Megaupload) make every possible maneuver not to be cached as file-based caching jeopardizes their business model. They lose the capability to monetize by - differentiated service, limited view time, inline ad insertion etc. DiViNetworks' solution leaves the whole control at the content provider's hand, and never sends a byte unless it has reached from the source.
- Non-HTTP: As Bytestream caching is agnostic to protocols, it also supports protocols different than plain HTTP, such as P2P-flavors, RTMP, Office app's etc.
- Live video: Traditional file caching cannot cache live content. Until the "file" is saved, no one watches it. With Bytestream caching one person's stream is a second person's history, even if they watch it only 50mSec later.
- Adaptive bit-rate (ABR): Lots of video today is streamed at ABR, which means that the video is comprised of 2-10 second fragments at different bit-rates, set according to bandwidth availability. Caching it in files is of course very challenging, and results in low hit-rate while for Bytestream caching it is transparent.
- Re-purposed content: The social Internet makes a lot of content re-purposing. Thus you can find the same image in various Facebook accounts, Picasa albums, and Google+ pages. File caching will identify the manifestations of the image as different, and will reach low hit-rate. Bytestream caching, again, doesn’t mind.
- Content lost due to inefficient storage: The typical user doesn’t consume "a file" anymore. The average length of a YouTube video is 4'12"whereas the average view time is about 40 seconds (see "Bytemobile: Video Optimization Increases Video Clip Viewing Time by 50%"- here). Caching the whole file means that most of the storage volume is spent on unvisited bytes. Bytestream caching stores only the most frequent bit-level fragments.