Bitcodin (YC S15) encodes video 100x faster, in Netflix-grade quality • Hacker News Board

Re: Bitcodin (YC S15) encodes video 100x faster, in Netflix-grade quality

#11

Post by mkagenius » Fri, Aug 14, 2015, 5:44 PM UTC

slederer wrote: post #4

dman wrote: post #3

Earlier quoted context omitted.

Could you provide some details on how you are encoding videos 100x faster?

Sure, basically other encode video second by second, we can do this in parallel on salable cloud infrastructure. We also released a blog-post on this, based on a writeup from our infrastructure partner Google Compute Engine here: https://www.bitcodin.com/blog/2015/08/how-bitcodin-utilizes-...

Hi Stefan,

Glad to see someone doing this. I am assuming the process goes like this:

1. Segment the video first - upload them to cloud.

2. Let Map-Reduce do the transcoding on the individual segments.

3. Fetch the segments back at a place.

4. Create the manifest file listing the individual segment.

Is that roughly correct?

Since you are doing this on GCE, I am assuming you are not using GPUs, only CPUs, right?

Transcoding aside, how are you serving the video, where do you store it and which CDN do you use?

Re: Bitcodin (YC S15) encodes video 100x faster, in Netflix-grade quality

#12

Post by slederer » Tue, Aug 18, 2015, 6:48 PM UTC

mkagenius wrote: post #11

slederer wrote: post #4

Earlier quoted context omitted.

Sure, basically other encode video second by second, we can do this in parallel on salable cloud infrastructure. We also released a blog-post on this, based on a writeup from our infrastructure partner Google Compute Engine here: https://www.bitcodin.com/blog/2015/08/how-bitcodin-utilizes-...

Hi Stefan, Glad to see someone doing this. I am assuming the process goes like this: 1. Segment the video first - upload them to cloud. 2. Let Map-Reduce do the transcoding on the individual segments. 3. Fetch the segments back at a place. 4. Create the manifest file listing the individual segment. Is that roughly correct? Since you are doing this on GCE, I am assuming you are not using GPUs, only CPUs, right? Transc…

yes, it's more or less how you described it, we just added some tweaks here and there to make it work in practice :-)

ad 1.) no, we first grab the video from the user and then segment it in the cloud

Yes, we just run CPU instances.

Typically the user transfers the file back, e.g. via our FTP-transfer, S3, GCS, or simple download as zip.

Re: Bitcodin (YC S15) encodes video 100x faster, in Netflix-grade quality

#13

Post by mkagenius » Fri, Aug 21, 2015, 9:15 AM UTC

slederer wrote: post #12

mkagenius wrote: post #11

Earlier quoted context omitted.

Hi Stefan, Glad to see someone doing this. I am assuming the process goes like this: 1. Segment the video first - upload them to cloud. 2. Let Map-Reduce do the transcoding on the individual segments. 3. Fetch the segments back at a place. 4. Create the manifest file listing the individual segment. Is that roughly correct? Since you are doing this on GCE, I am assuming you are not using GPUs, only CPUs, right? Transc…

yes, it's more or less how you described it, we just added some tweaks here and there to make it work in practice :-) ad 1.) no, we first grab the video from the user and then segment it in the cloud Yes, we just run CPU instances. Typically the user transfers the file back, e.g. via our FTP-transfer, S3, GCS, or simple download as zip.

Thanks for the reply. I tried it.

Just the resource allocation took 2 min and the transcoding was done in 15 seconds for a 4 min video, so looks good.

Regarding selling this thing, I don't know if there could be any good use case for normal people. New enterprises could be better target. Old enterprises too can benefit if their current price is more than what you are providing.