Anyone use a Cloud Service for Encoding?

Anything and everything to do with DCP-o-matic.
jgoewert
Posts: 1
Joined: Tue Jul 10, 2018 4:07 pm

Anyone use a Cloud Service for Encoding?

Post by jgoewert »

I've been thinking about doing this to have a way to speed up conversions, but haven't given it a try as I keep seeing the horror stories of misconfiguring something in the cloud service (like Google or Amazon) that would cause a huge spike in cost. I used the Google calculator and even checking out a 64 core processor instance would be able $3 to convert about 2 hours of video.

I have a Gigabit fiber connection so I get peeks around 800 to 900 mpbs. So, I think that would be the main limiter in terms of speed.

The plan would be to start up the instance and use the Linux ISO, open the ports for the encoding process and then have my Server feed it frames.

I keep thinking I am missing something, like a bandwidth cost or some weird per gig data transfer cost. Trying to figure out if there is some gotcha I'm just not seeing.
Carsten
Posts: 2804
Joined: Tue Apr 15, 2014 9:11 pm
Location: Germany

Re: Anyone use a Cloud Service for Encoding?

Post by Carsten »

I guess the cost benefit is there if you only do occasional conversions, and they need to be finished quickly. In the past, we usually came to the conclusion that the transfer bandwidth will limit the effective transcoding fps. Keep in mind, the encoding server receives uncompressed image series, so, downstream AND upstream need to be balanced in order to get a decent transcoding speed. What we do know is that when using local network encoding servers, with CPU core numbers in that range (32+), we need 10Gigabit networks to keep the encoding servers crunching away.

Cloud based may work if you can assume both source footage and final DCP being and remaining in the cloud, and close to the encoding server.


I guess it would be fun to try it. I remember that a long time ago Carl did try something similar for fun with a different cloud based computing service, but that was long ago I think and the results can not be compared with todays AWS.

- Carsten