Is it even possible, financially speaking, to keep adding storage? I mean, advertisements don’t even make a lot of money, is the indefinite growth of server storage even sustainable?
Or will they do what Twitch does with old content and just delete them?
They are not going to run out of addresses , and Alphabet (google & friends) who owns youtube has massive server farms , so storage is unlikely to ever be a problem.
Storage for YT is not like storage for your computer. The question to ask is not if Google has enough hard drive space to keep old videos, but what it costs Google to keep all YT videos available enough to meet demand.
First keep in mind that there isn’t just one giant server at YouTube. Everything is replicated onto many parallel servers. And enormous datasets that are too large for any one server are “sharded” across many. Perhaps it takes 1000 server clusters to store one copy of everything.
Now you have to parallelize copies of those 1000 so there are redundant servers that can scale up to meet viewer capacity. This is a server “grid.”
But only some videos are being watched millions of times today. Only those server nodes need 100x redundancy for scale. The long tail of less watched videos might barely need a single node to be kept available.
So there is a massive “head” of videos that need tremendous server capacity to be available enough, and a very long and thin “tail” of videos that don’t require much resources at all.
The “head” grows as YT’s overall audience gets bigger. It’s very resource hungry and is probably their main challenge.
The “tail” gets longer as the total library of videos grows. But the tail is thin and making it longer isn’t that expensive.
I’m sure there is also a threshold below which they will drop videos. Made over 1 year ago. More than 30 seconds long. Has never been viewed once. Auhthor account hasn’t been visited in a year either. Drop it. No one will ever know.
I have wondered the same from time to time, but I really don’t think so, every time tech becomes better and smaller (for the most part), it wouldn’t be crazy to think we can double or triple the current amount of normal storage we have in a few years, without compromising size at all, and Google has the funds to adapt these new tendencies first than most folks.
Once YouTube contains all the data in the universe, it will open its protective crust so as to scan itself, completing its thousand year task. Once their task is complete, they will ensure that no new information arises in the only way possible: By destroying the universe.
YouTube is known to reduce the quality of old videos. The resolution is often the same (e.g.1080p), but the image quality is way worse compared to when those videos were new. They’re probably doing it to reduce their storage cost.
They still keep the originals. They’re degrading the quality of older videos for bandwidth reasons, not storage.
Also, it’s not because of the price of bandwidth: It’s because they can use up so much bandwidth in a given region that they can cause slowdowns (e.g. hogging too much of what’s available).
There’s a physical limit to everything (I’m talking more land space here, they can’t just keep adding new drives for infinity), especially since a site like youtube will only get more expensive to run as time goes. I expect them to start deleting content that doesn’t make them money like videos with under 5 views that are over a year old relatively soon especially with the current economy. This would make sense for them and would free up a lot of their storage. Google doesn’t disclose the profit margins of youtube but I am pretty sure they are not very large especially now with 8K HDR videos being available
They’re already wiping inactive google accounts and all related content. It’s going to be problematic for old videos where the owners haven’t used the account in some time.
I have a friend that passed away past the limit, I’m going to need to make sure to archive all of his stuff or else it’ll all fall into the youtube void.
I got the memo but accounts with YouTube videos seem to be exempt from this. This is good because I uploaded some content a few years back and forgot the passwords.
I do not think do, but I read google is scraping for money from every paid user so maybe yes. Plus many people are leaving the platform to decentralized options.
Storage is cheap, especially at the corporate scale.
Make two simplifying assumptions: pretend that Google is paying consumer prices for storage, and pretend that Google doesn’t need to worry about data redundancy. In truth Google will pay a lot less than consumer prices, but they’ll also need more than 1 byte of storage for each byte of data they have, so for the sake of envelope math we can just pretend they cancel out.
Western Digital sells a 22TB HDD for $400. Seagate has a 20TB HDD for $310. I don’t like Seagate but I do like round numbers, so for simplicity we’ll call it $300 for 20TB. This works out to $15/TB. According to wikipedia, Youtube had just under $29b of revenue in 2021. If youtube spend just $100m of that — 0.34% — they’d be able to buy 6,666,666 of those hard drives. In a single year. That’s 6,666,666x20TB = 133,333,333 TB of storage, also known as 133^note^ ^1^ exabytes.
That’s a lot of storage. A quick search tells me that youtube’s compression for 4k/25fps is 45Mbps, which is about 5.5 megabytes/s. That’s 768,722 years of 4k video content. All paid for with 0.34% of youtube’s annual revenue.
Note 1: Note that I am using SI units here. If you want to use 1024^n^ for data names, then the SI prefixes aren’t correct. It’d be 115 exbibytes instead.
I know you are saying Google doesn’t have to worry about redundancy to simplify the math but I think that makes it completely useless.
Redundancy is not just about having another copy incase of data loss but more importantly for enterprises redundancy allows for more throughput. If each video was on a single hard drive the site would not be able to function as even the fastest multi actuator hard drive can only do 524 MB/s in a perfect vacuum.
It’s useless for answering a questions that wasn’t asked, sure. But I didn’t pretend to answer that question. What it is useful for is answering the topic question. You know, the whole damn point?
How much of a factor off do you think the estimate is? You think they need three drives of redundancy each? Ten? Chances are they’re paying half (or less) for storage drives compared to retail pricing. The estimate on what they could get with $100m was also 134 EB, a mind boggling sum of storage. I wouldn’t be surprised if they’re using up on the order of 1 EB/year in needed storage. There’s also a lot more room in their budget than 0.34%.
The point is to get a quick and simple estimate to show that there really will not be a problem in Google acquiring sufficient storage. If you want a very accurate estimate of their costs you’ll need data that we do not have. I was not aiming to get a highly accurate estimate of their costs. I made this clear, right from the beginning.
If each video was on a single hard drive the site would not be able to function as even the fastest multi actuator hard drive can only do 524 MB/s in a perfect vacuum.
The most popular videos are all going to be kept in RAM, they don’t read them all off disk with every single view request. If you wanted a comment going over the finer details of server architecture, you shouldn’t have looked at the one saying it was doing back of the envelope math on storage costs only, eh?
I wouldn’t assume Googe pays less for storage. They need to pay for land use in many countries, power usage, redundancy and the staff that manages all of it.
They also need powerful servers with fast caching storage and a lot of RAM. They also need to pay for the bandwidth.
As far as I know, they save multiple copies of each video in all resolutions they serve. So an 8K video will also have 4K + 1440p + 1980p + 720p + 480p + 240p + 144p Possibly also 60Hz and 30Hz for some of them and also HDR versions.
You have to add all that to the cost per TB. Finally, there is the question of how much additional storage they need per year, 100 PByr? Presumably also increasing yearly?
Like LetMeEatCake said, storage is cheap, cheaper than CPU power. I don’t know for sure, but if you have a video that will be played 1000 times, what would be better? Transcode 1000 times or store 8 versions of the video?
I would imagine they store the highest available quality only by default and do on the fly transcoding until a certain threshold of views per time is reached. At this point they would then store the transcoded versions as well.
For videos with a lot of views it only makes sense to store the transcoded version as, like you say, storage is cheap. But fact is that the vast majority of uploaded videos get <1k views and for those it probably would make more sense to transcode them otf when needed.
I’d say there’s a significant amount of storage available for cache. On the first transcoding the video is saved to cache so that the cached version can be streamed in the future. After enough time lapses or the cache becomes full, the cached video is deleted
I wasn’t calculating server costs, just raw storage. Google is not buying hard drives at retail prices. I wouldn’t be surprised if they’re paying as little as 50% of the retail price to buy at volume.
All of what you say is true but the purpose was to get a back of the envelope estimation to show that the cost of storage is not a truly limiting factor for a company like youtube. My point was to answer the question.
With the level of compression youtube uses, the storage costs of everything below 4k is substantially lower than 4k by itself: for back of envelope purposes we can just ignore those resolutions.
Google has said they’ll start deleting long inactive accounts. I’m guessing they’ll go for unlisted next. Eventually they’ll try and prune their catalog to not have tons of unprofitable videos.
Google datacenters are global. They store petabytes of data in each and are constantly evolving to Google Cloud customer needs and their own, meaning they are always expanding their storage network with new servers and drives.
In addition to the hardware expansion, YouTube engineers are experimenting with ways to encode videos in much more efficient formats, such as AV1. Basically, encoding is how a video is stored. The engineers are trying new standards to retain original video quality in much smaller file sizes, leading to more video storage capacity without the need to upgrade servers as quickly.
Add comment