#Performance rendering inside Docker in GCP

4 messages · Page 1 of 1 (latest)

steel shard
#

Hey guys, I'm following your example of running the remotion renderer in docker. I'm deploying it in GCP.
I assigned it plenty of RAM and a couple of vCPUs.
When I render short videos, the time it takes to render is somewhat acceptable, but it gets insanely long for a 2 minute long video where it's just two video files one after the other.

My composition is essentially the Timeline composition that I'm re-using.

When I look at my machine, it's barely using any resources.
It's taking between 200 to 500ms per frame being rendered.
My renderer concurrency setting is set to "75%"

Could I get some insights on what could be the bottleneck, what can I do to speed this up?

steel shard
#

When I check how long each frame takes to generate, a lot of the frames show 500ms, which is an odd clean number, indicating maybe its hanging up and hitting a timeout during the frame gen

remote cliff
steel shard
#

I havent added the verbose logs. I'm logging my frame times but it doesnt tell me why its slow. Note that I improved the situation a lot by using a number for concurrency instead of "75%", which didnt seem to do anything since we are running inside a node in GCP with variable vCPUs