How Instagram made its video infrastructure 94 percent more efficient
How Instagram made its video infrastructure 94 percent more efficient
People love creating and sharing videos on Instagram. However, Instagram’s increasing user growth has put pressure on Meta’s video processing infrastructure. In early 2021, our projections showed that within 12 months we would not have enough capacity to provide video uploads for everyone. That same year, Instagram’s engineers developed a way to reduce the compute workload of Instagram’s less-watched video encodings by 94 percent — all while using existing hardware.
With more resources available, we were able to increase the overall watch time coverage of advanced encodings by 33 percent. This means that more people who watch video on Instagram now get to experience clearer video that plays more smoothly. This is especially beneficial for people in countries that have slower internet connections.
With more resources available, we were able to increase the overall watch time coverage of advanced encodings by 33 percent. This means that more users who watch video on Instagram now get to experience clearer video that plays smoother. This is especially beneficial for users in countries that have slower internet connections.
How it works:
When a video is uploaded to Instagram, it gets encoded into different versions — each with different characteristics. Instagram’s engineers found they could repurpose one type of video encoding to help generate others. Doing this significantly reduces the compute resources spent on less-watched video encodings, while also allowing Instagram to deliver clearer videos that play more smoothly.
Why it matters:
In our constant quest for efficiency, making better use of the machines that deliver our service was a top priority. Moving forward, this optimization will allow Instagram’s infrastructure to scale to meet the needs of our more than 2 billion monthly active users.
Take a deeper dive:
Learn more about how Instagram’s engineers reduced compute time spent producing less-watched encodings by 94 percent.
.png)
Comments
Post a Comment