Instagram (and Meta) Throttle Video Quality as Views Go Down (theverge.com) 49
An anonymous reader shared this report from the Verge:
Ever wondered why some of your Instagram videos tend to look blurry, while others are crisp and sharp? It's because, on Instagram, the quality of your video apparently depends on how many views it's getting.
Here's part of Mosseri's explanation, from the video, which was reposted by a Threads user today. "In general, we want to show the highest-quality video we can ... But if something isn't watched for a long time — because the vast majority of views are in the beginning — we will move to a lower quality video. And then if it's watched again a lot then we'll re-render the higher quality video...."
The shift in quality "isn't huge," Mosseri said in response to another Threads user, who'd asked if that approach disadvantaged smaller creators. That's "the right concern," he told them, but said people interact with videos based on its content, not its quality. That's consistent with how Meta has described its approach before... Meta wrote in a blog [post] that in order to conserve computing resources for the relatively few, most watched videos, it gives fresh uploads the fastest, most basic encoding. After a video "gets sufficiently high watch time," it receives a more robust encoding pass.
"It works at an aggregate level, not an individual viewer level," Mosseri wrote later on Threads. "We bias to higher quality (more CPU intensive encoding and more expensive storage for bigger files) for creators who drive more views. It's not a binary theshhold, but rather a sliding scale."
Here's part of Mosseri's explanation, from the video, which was reposted by a Threads user today. "In general, we want to show the highest-quality video we can ... But if something isn't watched for a long time — because the vast majority of views are in the beginning — we will move to a lower quality video. And then if it's watched again a lot then we'll re-render the higher quality video...."
The shift in quality "isn't huge," Mosseri said in response to another Threads user, who'd asked if that approach disadvantaged smaller creators. That's "the right concern," he told them, but said people interact with videos based on its content, not its quality. That's consistent with how Meta has described its approach before... Meta wrote in a blog [post] that in order to conserve computing resources for the relatively few, most watched videos, it gives fresh uploads the fastest, most basic encoding. After a video "gets sufficiently high watch time," it receives a more robust encoding pass.
"It works at an aggregate level, not an individual viewer level," Mosseri wrote later on Threads. "We bias to higher quality (more CPU intensive encoding and more expensive storage for bigger files) for creators who drive more views. It's not a binary theshhold, but rather a sliding scale."
Hmmm? (Score:5, Interesting)
There has to be a "highest res" data file stored somewhere to generate a highest res video. So how is there disk space saved?
And once they have that highest res file already on disk (they'd have to) then what cpu are they using to send it to a viewer?
I can see how lower res would save bandwidth but the rest of the summary makes no sense. There's something missing in that explanation.
There's also a psychological component and catch-22 here. If a video looks bad then it will get fewer viewers sharing it with others. Don't they want videos shared?
It feels very bean countery. But dumb beans not smart beans.
Re:Hmmm? (Score:5, Informative)
Because when you're operating at that scale, there isn't just "one" copy of a file. There is a CDN with replcas all over the world. The original full-quality video that isn't served to end-users would only need to be stored essentially once (plus whatever level of redundancy) - and when it comes time to re-render, only that one copy is needed, and then the re-rendered file will be pushed out to the hundreds/thousands of CDNs.
Re: (Score:2)
Ok, that makes more sense. Although an unpopular video won't be on every cdn node. If it gets 10 views during a cdn's file cache life cycle time then worst case is 10 copies are floating around for a few hours. If it's more popular then it deserves full res anyway as they already do.
This still looks like they're chasing after the trivial end tail of the bell curve counting beans while hurting video vitality.
Almost there... (Score:2)
Although an unpopular video won't be on every cdn node.
That's almost litteraly what Mosseri tried to explain about Instagram video.
Except that he went into way too many tiny details on how the cache evistion logic is handled (the popularity of both a video and a channel in general are factors, in adition to more classic metrics like "cache hits" as you might be used with the cache on your desktop machine) without dwelling much on the other reason why it works this way (basically "latency". People are used that whenever the tap on something it immediately starts
Re: (Score:2)
I agree with the concept of what you're saying but the details are getting lost.
If I have an unpopular video which gets 10 hits a week, who cares if it slow loads straight from the origin node for 10 people? If they have to convert and copy it to the local cdn anyway, that will actually take longer than just sending it as-is. And for 1-2 people a day? So what? It's totally not worth the effort to down res it and then send that.
This is all trivial tail end stuff. I doubt it's even measurable disk or ban
Wait time. (Score:2)
If I have an unpopular video which gets 10 hits a week, who cares if it slow loads straight from the origin node for 10 people?
Those 10 people are going to complain, because the video doesn't start playing the instant their finger approaches the touch screen.
Remember that Instagram is a network whose initial shtick was to start uploading you photo in the background while you're still playing with filter, so by the time you've finished selecting them, it's already there and it gets instantly posted the moment you tap the button.
I bet the amount of users who want absolute instantaneous response is even more disproportionate on Instag
Re: (Score:2)
The problem is you can't predict if a video is just going to be seen by 10 people or 10 million. And you really don't want to figure out that say, 10 people are seeing it, send it to the CDNs only to have it sit there because that's all.
Chances are, you quickly produce a low quality version, and then send it to a CDN of the first request. The video is low quality and can be sent basically instantly. If a second person requests it, their CDN will pull the video from the first CDN basically instantly as well
Re: (Score:2)
They're saying the opposite. Their claim (not mine) is that the cpu requirements to deal with rarely seen videos is very high as well as storage.
Again, since we don't have their metrics we're just shooting in the dark here.
They likely already have regional origins for all their data so the hires file is already "nearby". If I was them I'd send the hires directly on first hit in a region. If the viewer clicked the share button or got more than x% through the video then I could worry about making a low res
Re: (Score:2)
Makes sense because if videos aren't watched, they sure wouldn't save much on bandwidth with their scheme.
Re: (Score:1)
Because when you're operating at that scale, there isn't just "one" copy of a file. There is a CDN with replcas all over the world. The original full-quality video that isn't served to end-users would only need to be stored essentially once (plus whatever level of redundancy) - and when it comes time to re-render, only that one copy is needed, and then the re-rendered file will be pushed out to the hundreds/thousands of CDNs.
Also known at this profit scale, as the cost of doing business? Let me break out my pocket violin for all the “high-def” we get these days, with broadcasts still streaming in the same 720 resolution they offered a decade ago that got them that “HD” badge of (minimal effort) honor.
Stay tuned for that Jan 1 new fiscal year announcement-excuse when they start offering a “premium” paid membership that allows you “unlimited” access to the streaming quality you had
Re: (Score:2)
Suppose your users are willing to watch a certain base quality of video content. If you offer any file with an enhanced quality, then the users would still watch it with the base quality if that was offered instead without a choice. So you're literally wasting bandwith and computation by offering the hig
Re: Hmmm? (Score:2)
Does it even make sense to CDN something unless its viewed a lot? Makes me wonder if they adjust that based on views as well....
Re: Hmmm? (Score:2)
Platforms like insta have multiple levels of storage. It's not really about raw disk space but rather different types of space. There are multiple levels of content cache on these massive content delivery architectures, and what he's describing is the strategy to optimise resources. (And like you noted, bandwidth matters too.)
Re: (Score:2)
There has to be a "highest res" data file stored somewhere to generate a highest res video. So how is there disk space saved?
You may want to research CDNs.
Re: (Score:2)
Re: (Score:1)
Re-encoding multiple times with different quality levels is done to manage a number of challenges, primarily bandwidth costs. CDNs have multiple copies of all files they distribute so they can be served to a given client from the closest node at the most appropriate quality level. A large CDN can easily have thousands of nodes, and deciding how many copies of each file are stored on which node with what quality level and delivered to which client is a really, really complex problem that is not done manually
Re: (Score:1)
Yea see this is where my question "do you know how the internet works" comes in.
Anyone who knows how a CDN works will not need the details you're asking for. Yea the article is not heavy on details, and it reads like it was written by a journalist without a lot of background knowledge on the issue. Bandwidth costs, storage costs, maintenance costs, staffing costs, electricity costs, these all factor in to the algorithm that powers their server's decisions on what video quality to store where and for how lon
Re: (Score:1)
I didn't say it's not about storage, I said it's about a number of factors, all of which CDNs try to minimize.
I was responding to the initial claim that they have "highest res" version and that ought to be enough. That's it. That's the dumb assertion that I responded to. I didn't make any categorical statements that it's about this or that, nor did I say it's not about storage.
Your insistence in attempting to strawman my statements into something I didn't say, and then prove that to be wrong leads me to bel
Re: (Score:1)
What a fucking dumb question. Do you even know how the internet works?
You obviously don't or you wouldn't be generalizing that the entire Internet works one way. This, is also clearly not about the Internet but storage space. Why else would they be re-encoding multiples times to high quality?
For those justifying CDNs and the burden of re-encoding multiple times, this is called the cost of doing business.. Not sure when we forgot that.
If they need operating revenue, maybe they should stop paying Mass Narcissism so much, and we should stop forgetting how social media works.
Re: (Score:2)
I was thinking the same thing but I realized they are referring to the storage space on the thousands of CDN edge servers spread throughout the world, not the storage space of the original video.
The lower resolution video is also saving the network bandwidth to those edge servers.
Hmm... (Score:2)
Ever wondered why some of your Instagram videos tend to look blurry, while others are crisp and sharp? It's because, on Instagram, the quality of your video apparently depends on how many views it's getting.
Sounds like a negative feedback loop.
And then if it's watched again a lot then we'll re-render the higher quality video...."
Are more or less -- or more correctly, fewer -- people going to watch, or continue to watch, fuzzy videos?
Weird stuff (Score:3)
Re: (Score:3)
Re: (Score:2)
Maybe just maybe creators should start hostng the vudeos on platforms that compress lees and just link to said videos from insta et all.
You could be forgiven for thinking so. But in reality, Meta (.et al) have configured their services to show users fewer external links and more internally hosted content, because they want to keep as many eyeballs on their platform/app as possible and stop people navigating away to other services. The result of which is that if you host a video (or anything else) externally and provide a link, far less people will be shown what you've posted vs if you had uploaded it to their internal servers.
Re: (Score:2)
Re: (Score:2)
Tech journos and IT leads cannot even use the right terms, no, they are not lowering "quality", they are pushing videos with lower "bitrate". Meta apparently wants to save every thousandth of a penny on their bandwidth costs.
We mock the savings efforts (perhaps rightfully so), but I do envision a network engineer with some level of integrity and talent coming forward with this proposal, who probably found a cost savings a LOT larger than we’re mocking.
If they cared enough to mitigate the mockery, they could simply show us those savings.
Re: (Score:2)
no, they are not lowering "quality", they are pushing videos with lower "bitrate"
When you lower the "bitrate" you also reduce the "quality" so there is nothing factually incorrect about their statement. Is it the most accurate term they could use, no. Is it one that people will be familiar with, yes. Is it also one that people who actually understand which issues are relevant will complain about, no.
Re: (Score:2)
Sigh. (Score:5, Funny)
"Nobody is watching us!"
"I know, let's make it look even worse!"
"Great idea!"
Re: (Score:2)
Exactly. However if you cannot think beyond the next quarter, it makes perfect sense. MBAs at work ...
Now I know who to curse (Score:4, Insightful)
Most people here are saying it is a totally reasonable decision because, CDN.
My first reaction though was, thank god I don't use them for a publishing platform.
CDNs are there to solve this problem, guys.
When you publish video you are expecting ads to be paying for it, and maybe you are expecting to make money off it.
But if there is a hurdle to gaining viewers because your files are intentionally not being served at their best resolution, that means the money is going to the platform owner instead of paying for the resources needed for publication.
So the platform owners are optimizing CDN deployment strategies to optimize their income, and not to optimize Content Distribution from the author's perspective.
I don't see that as reasonable, unless they made this behavior very clear from the outset.
Personally I don't even have the greatest vision but still often click wondering isn't there a better resolution? Now I know why and who to curse when things look so blurry.
That seems ass-backwards (Score:2)
Typical MBA-thinking.
Low on compute? (Score:2)
Self-fulfilling prophecies of success or failure. (Score:3)
Wanna get back at Zuckerberg? Here's an idea (Score:2)
if it's watched again a lot then we'll re-render the higher quality video...."
FInd old videos, automatically view them a bunch of times (possibly a distributed kind of attack) until Facebook wastes resources and money re-encoding them for nothing.
Hmm...