HTML5 Video buffered attribute features

Question

I am designing a custom HTML5 video player. Thus, it will have its own custom slider to mimic the video progress, so I need to understand the entire buffering shebang of a HTML5 video.

I came across this article: Video Buffering. It says that the buffered object consists of several time ranges in linear order of start time. But I couldn't find out the following:

1. Say the video starts. It continues upto 1:45 on its own (occasionally stalling perhaps, waiting for further data), after which I suddenly jump to 32:45. Now after some time, if I jump back to 1:27 (within the time range initially loaded and played through, before I made the jump), will it start playing immediately as it was already loaded before? Or is it that since I made a jump, that portion is lost and will have to be fetched again? Either way, is the behavior consistent for all such scenarios?

2. Say I make 5 or 6 such jumps, each time waiting for a few seconds for some data to load after the jump. Does that mean the buffered object will have all those time ranges stored? Or might some get lost? Is it a stack kind of thing, where the earlier ranges will get popped off as more ranges get loaded due to further jumps?

3. Will checking whether the buffered object has one time range starting at 0 (forget live streaming) and ending at the video duration length ensure that the entire video resource has been loaded fully? If not, is there some way to know that the entire video has been downloaded, and any portion is seekable, from which video can play continuously upto end without a moment's delay?

The W3C specs are not very clear on this, and I also can't find a suitably large (say more than an hour) remote video resource to test.

Answer 1

How video is buffered is browser implementation-dependent and therefore may vary from browser to browser.

Various browsers can use different factors to determine to keep or to discard a part of the buffer. Old segments, disk space, memory, and performance are typical factors.

The only way to know is to "see" what the browser has or is loading.

For example - in Chrome I played a few seconds then I skipped to about 30 seconds and you can see that it starts to load another part starting from that position.

(The buffer also seem to be bounded to key-frames so it is possible to decode the n-frames in that buffer. This means the buffer can start to load data a little before the actual position).

I supplied a demo video about 1 minute long - however, this is not long enough to do proper testing. Free free to supply video links that contain longer video (or please share if you want me to update the demo with this).

The main function will iterate through the buffered object on the video element. It will render all parts that exist to the canvas right below the video showing in red.

You can click (bit not drag) on this viewer to move the video to different positions.

/// buffer viewer loop (updates about every 2nd frame)
function loop() {

    var b = vid.buffered,  /// get buffer object
        i = b.length,      /// counter for loop
        w = canvas.width,  /// cache canvas width and height
        h = canvas.height,
        vl = vid.duration, /// total video duration in seconds
        x1, x2;            /// buffer segment mark positions

    /// clear canvas with black
    ctx.fillStyle = '#000';
    ctx.fillRect(0, 0, w, h);

    /// red color for loaded buffer(s)
    ctx.fillStyle = '#d00';

    /// iterate through buffers
    while (i--) {
        x1 = b.start(i) / vl * w;
        x2 = b.end(i) / vl * w;
        ctx.fillRect(x1, 0, x2 - x1, h);
    }

    /// draw info
    ctx.fillStyle = '#fff';

    ctx.textBaseline = 'top';
    ctx.textAlign = 'left';
    ctx.fillText(vid.currentTime.toFixed(1), 4, 4);

    ctx.textAlign = 'right';
    ctx.fillText(vl.toFixed(1), w - 4, 4);

    /// draw cursor for position
    x1 = vid.currentTime / vl * w;

    ctx.beginPath();
    ctx.arc(x1, h * 0.5, 7, 0, 2 * Math.PI);
    ctx.fill();

    setTimeout(loop, 29);
}