If 10 million people are watching youtube does that mean there are 10m harddrive

[Beamin]

When you watch a video on youtube is there a actual hard drive at the server end that spins up seeks the video on the disk then plays back that video with that particular hard drive staying in use by you until the video is over? How do they do this if say 100 million people are watching it at once they don't have millions of hard drives or do they? And if 65 years of videos are uploaded each day how do they keep up with the memory requirements? How many gb/tb are in each sever farm? It seems like they would end up spending huge amounts of money to store videos many of which are shit and will get a handful of views each year and really just take up memory and cause delays in indexing all these videos.

[martin1454]

Well, there is not 100 million harddives for 100 million people - 1 harddrive can read multiple videos off to multiple users, so it is probaly something like 1 HDD = 10-100 users.

[DannyTheGhost]

Well, i somewhere read article about cloud servers holding data
First of all, who said that 100 million people will watch 100 million different videos?
So, for comfort reasons, data which is 'popular' (that is frequently requested) will be moved from high-capacity low-speed hard drives to low-capacity high-speed drives so more people will accept it fast
In case when data is stored for long time it will be moved back to those 10TB+ hard drives
Storage is not the problem when we talk about servers - it's its bandwidth

[AndyC_772]

Popular videos will be held on SSDs, or even in a RAM cache, for speed and power reasons. They'll only get moved onto hard drives once demand has fallen.

[EEVblog]

It seems like they would end up spending huge amounts of money to store videos many of which are shit and will get a handful of views each year and really just take up memory and cause delays in indexing all these videos.

Yes, and no.
The most viewed videos would get priority space and caching at the local CDN's

[Electro Detective]

One should ask about all the server wastage on the amount of garbage commercials/mega annoyances on Youtube,
stuff no normal consumer would have use for or ever buy,
much less understand wtf it is   

commercials clearly produced and performed by hoards of insane asylum escapees and or ice indulging coffeeholics.     


and then there's the annoying commercials that must be paying Youtube a fortune to bully and jack our browsers, like WIX, Grammarly, UAP, and others,
BS that can easily pollute a short video with heaps of ads, popups and browser freeze ups and crashes.

I suspect the problem is frustrated viewers using ad blockers,
and Youtube kicking back harder, which hits the legit 'cop it sweet' non ad block users like me   

[bloguetronica]

...
I suspect the problem is frustrated viewers using ad blockers,
and Youtube kicking back harder, which hits the legit 'cop it sweet' non ad block users like me   

The fault is on YouTube and not on the ad-block users. Ad-block users are as legal as any other users, since they don't wan't their PC's plagued by virus. On my smartphone, google ads often suggests me ads to spyware, bloatware or virus on legit websites. On the PC, ads are no less dubious.

Kind regards, Samuel Lourenço

[GlennSprigg]

NOT just speaking of 'Youtube', but in general....
A lot of lay people don't grasp how big 'Google' is !!  They imagine a 'server/servers' somewhere,
when there are huge buildings all over the globe, each filled with thousands of computers/servers !!

Not to mention Google's ridiculously huge & complex array of Databases, as a result of being, (in a
very simplistic analogy!), like a huge global 'Index' for 'Yellow-Pages' phone-books that was missing.
It still amazes me when I do a 'Search' for a phrase/image etc, & finds 3,600,000 results in say .3 Sec !!!
(Although yes, that includes smart analyzing/preempting of what you type, WHILE you are typing it).

Of course some of their data is updated almost in real time, or up to 24 hrs later to update new Domain
Name Server info for new sites etc, but 'generally', (I think??), their Systems still re-download the
ENTIRE Worlds Internet, at least once a week, to update all their servers/databases !!!   

No more "10 386-PC's daisy chained together in a shed" !!!   

[tooki]

Yep. Google’s infrastructure is incredible.

The stats for YouTube alone are amazing. This was a Facebook post I made back in December 2017:

Today’s tech factoids:
• Even though I literally can’t imagine an Internet without it now, YouTube wasn’t launched until 2005. That’s *after* MySpace and Facebook!
• Just two years later, in 2007, YouTube alone consumed as much bandwidth as the *entire Internet in the year 2000*.
• Today, visitors watch a billion hours of video on YouTube every day.
• By early 2017, 400 hours of video gets uploaded every *second*. That’s 66 YEARS of video added each day!

No matter what one thinks of Google and YouTube’s business practices, the scale of YouTube and the infrastructure to run it is just mind-blowing.

The last item just blows my mind. The storage (redundant and distributed worldwide!), the processing for re-encoding...

[tooki]

Also, do you guys remember this classic??? 

https://youtu.be/SXmv8quf_xM

I want some of that “tracer T” action...

[GlennSprigg]

'tooki' Ha !!!   
Luckily his video was made when he was an ancient 12 year-old or something.
As when he was only 8 years old, he knew NOTHING man!!!!!!!!   

[tooki]

As the comments said... if that's the "next gen" of hackers (see his username), then IT security folks will soon be out of business!

[james_s]

One should ask about all the server wastage on the amount of garbage commercials/mega annoyances on Youtube,
stuff no normal consumer would have use for or ever buy,
much less understand wtf it is   

commercials clearly produced and performed by hoards of insane asylum escapees and or ice indulging coffeeholics.     


and then there's the annoying commercials that must be paying Youtube a fortune to bully and jack our browsers, like WIX, Grammarly, UAP, and others,
BS that can easily pollute a short video with heaps of ads, popups and browser freeze ups and crashes.

I suspect the problem is frustrated viewers using ad blockers,
and Youtube kicking back harder, which hits the legit 'cop it sweet' non ad block users like me   

YouTube is unusable without an ad-blocker, I don't even bother to try.

[mzzj]

NOT just speaking of 'Youtube', but in general....
A lot of lay people don't grasp how big 'Google' is !!  They imagine a 'server/servers' somewhere,
when there are huge buildings all over the globe, each filled with thousands of computers/servers !!

Not to mention Google's ridiculously huge & complex array of Databases, as a result of being, (in a
very simplistic analogy!), like a huge global 'Index' for 'Yellow-Pages' phone-books that was missing.
It still amazes me when I do a 'Search' for a phrase/image etc, & finds 3,600,000 results in say .3 Sec !!!
(Although yes, that includes smart analyzing/preempting of what you type, WHILE you are typing it).

Of course some of their data is updated almost in real time, or up to 24 hrs later to update new Domain
Name Server info for new sites etc, but 'generally', (I think??), their Systems still re-download the
ENTIRE Worlds Internet, at least once a week, to update all their servers/databases !!!   

No more "10 386-PC's daisy chained together in a shed" !!!   

Back in the days I remember reading from somewhere that either google or Altavista used to transfer the search index (on tapes?) weekly across the atlantic on aeroplane as internet capacity at that time was too limited..

[TomS_]

YouTube is unusable without an ad-blocker, I don't even bother to try.

I wouldnt say its unusable, thats a bit too far. Its perfectly usable.

There are types of ads which I find particularly annoying though, namely the ones that are scheduled in to the middle of a video. Sometimes multiples of.

Im quite happy watching ads at the start of a video, even if you cant skip them, if it means the rest of the video is uninterrupted - I mean, they have to make money to keep providing that service somehow. But interrupt the video and you wreck the experience. Sometimes, very well made, high quality content videos are ruined by multiple ads playing throughout.

[LapTop006]

YouTube is unusable without an ad-blocker, I don't even bother to try.

I wouldnt say its unusable, thats a bit too far. Its perfectly usable.

There are types of ads which I find particularly annoying though, namely the ones that are scheduled in to the middle of a video. Sometimes multiples of.

Im quite happy watching ads at the start of a video, even if you cant skip them, if it means the rest of the video is uninterrupted - I mean, they have to make money to keep providing that service somehow. But interrupt the video and you wreck the experience. Sometimes, very well made, high quality content videos are ruined by multiple ads playing throughout.

Whereas I simply pay the money for YouTube Red/Premium/Whatever it's called this week. The only times I see ads on YouTube is the about once I year I have to re-login, or in an incognito tab.

Also, as I've mentioned elsewhere, I work for Google, for those who want to know more I suggest the borg paper.

[mariush]

To answer the original question...

Google has multiple datacenters (places on the planet where they have servers).

Youtube uses servers located in most (if not all) of Google's datacenters.
Some servers are only used to process database requests.
Some may not have hard drives but instead have loads of RAM, like for example 256-512 GB of RAM. For example, such servers would be used to serve the thumbnails, video previews, user logos, static close captioning ... stuff that doesn't change often. 
Instead of requesting a small file from a spinning disc, with all the latency and IO involved, one of these servers with lots of RAM keeps copies of these small files in RAM and serves them directly from RAM.

Some servers may serve videos but not actually have the videos physically stored on them ... think of a big NAS with hundreds of hard drives, which is connected to the datacenter network with multiple 40-100gbps network cards 

A single dedicated server with a 10gbps network card can probably only serve around 50-100 videos at any single time, in order to guarantee let's say minimum 10-30 mbps download speed - so when you hit play on a video, the server asks the big nas machine for the video, the nas machine sends it to the server at 10 gbps speeds and the server holds it in RAM and then sends it to the user at a consistent speed.

Google saves the original videos and also creates multiple versions of the video at various resolutions and with various codecs, to support various devices (apple portables, various TVs etc).
Depending on popularity of a video, it will be stored in more datacenters... if the video is not so popular, you may find out it's downloaded from a region that's further away from you (across an ocean for example)... same if the video is very old and accessed rarely... they'll gradually drop such videos from some datacenters, or reduce the number of resolutions available in datacenters with fewer servers (ex max 720p instead of 1080p or 4k)

[rstofer]

A 1080p YouTube video will transfer about 50 MB per minute.  Call it 400 Mb/minute and realize that I can download at 350 Mb/SECOND.  In other words, they can blast the entire 60 minute movie to my PC in about 68 seconds.  The duty cycle is about 1 in 60.  1 second of transfer for 10 seconds of movie - at 1080p.

At 144p the transfer is about 1.3 MB per minute.  A 60 minute movie would be 78 MB or 624 Mb and I could slurp that up in a couple of seconds.

Internet bandwidth might be a limitation and certainly individual download rates will impact what is displayed (is dial-up dead?) but it is always the receiver that is throttling the transfer.  Yes, the server will throttle, particularly when loaded but more that likely, throttling occurs on my end or on the small fiber optic coming to the neighborhood.

My 350 Mb transfer rate is miniscule compared to the large pipes leaving the server farms.

SSDs are up in the GB/second range for server applications.  That 1080p movie will take, at most, 3 seconds of SSD drive access.  A single drive can handle about 1200 such movies concurrently.  For the 144p movie, the drive time would be a couple of dozen milliseconds and a single drive can deliver over 100,000 videos concurrently.

Bandwidth is everything!  That's why the ISPs want these huge data sources to pay more toward building up the infrastructure.

ETA:  All number are 'more or less'.  The details aren't as important as the capabilities of large pipes and large SSDs.  RAM is faster, HDDs are slower (a lot slower) but under any condition, it isn't one drive per movie.

[rstofer]

Back in the days I remember reading from somewhere that either google or Altavista used to transfer the search index (on tapes?) weekly across the atlantic on aeroplane as internet capacity at that time was too limited..

Never underestimate the bandwidth of a station wagon full of floppies.

[GlennSprigg]

Back in the days I remember reading from somewhere that either google or Altavista used to transfer the search index (on tapes?) weekly across the atlantic on aeroplane as internet capacity at that time was too limited..

Never underestimate the bandwidth of a station wagon full of floppies.

Yea... that's amazing. Speaking of 'floppies', I remember only so many years back, when Computer Magazines
has a single 1.44 mb disk taped to the cover. Was so excited to see what it contained !!!   
And the 'excitement' at hearing a '300-baud' 'modem' scream/blip/screech upon dial-up, was too much ! 
But that was when real tight coding was done due to RAM limitations etc.  An 'App' then, maybe '48k', did the
equivalent to a lot of apps today from 10mb to 80mb. I know how it 'works' today, and why... but that accounts
for 95% of the 'data' on the 'net' these days.

[Richard Crowley]

Over here in the Oregon Territories, Google takes advantage of the power from one of the Columbia River dams just a couple miles upstream, AND the cold water from the river to keep the servers cool.  AND the location of a major optical fiber link in the interweb for connectivity.  One of the Google server farms in The Dalles, Oregon:



https://goo.gl/maps/NawMBTtp6GVcPETK9

[Berni]

Yep its taken care of by CDN servers all over ths world.

Infact to get around the bandwith issue Google offers free CDN servers to the internet providers. They give the ISP a few big rack mount machines that they can stick in there own racks and connect them into there own switches. These mini CDN servers fetch the popular content from Google and keeps it cached localy. Then when a internet subscriber wants to watch any of this content it gets intercepted and served to the user directly trough the ISPs switch. The data doesnt even go trough the internet since it never needs to exit the ISPs own network.

The ISP is happy because it saves bandwith on there expensive terrabit links to the internet exhanges, so they will gladly swallow the cost of powering and cooling these small servers. Google is also happy becuase it saves bandwith on there own expensive internet links, saves server load and saves them on the power bill, all for the cost of a one time investment of handing out free servers.

Something similar is done on large LAN partys. There people download games of tens of GB in size from Steam. So to keep things fast they build there own Steam cache server and plug it into there massive LAN. All traffic from the Steam download CDN is passed trough a bank of these cache servers. They hold on to the data and serve it from memory when needed wouthout going into the internet at all. Because many people are downloading tge same game this saves a ton of internet bandwith and allows for somthing like 100 people downloading a game at 1Gbit/s EACH.

[james_s]

Never underestimate the bandwidth of a station wagon full of floppies.

Hmm let's see. My station wagon is a 1990 so right around the start of the 3.5" floppy era has a stated maximum load capacity of 1,040lbs. Subtract ~180lbs for the driver and that leaves 860lbs for cargo. A quick search came up with 19g for the weight of a typical 3.5" floppy, so 860lbs is about 390kg and let's figure about 5kg for boxes to carry the disks in so 385kg, I get about 20,263 disks for a little more than 29GB of data assuming the disks are all neatly filled.

I suppose that's not bad throughput even for a drive of several hours although it ignores the time it takes to transfer the data to and from the disks and the latency is wicked.

[Berni]

Well back in the era of 3.5" floppies the best  internet connection you could have likely gotten access to at home is a ISDN line. With the blazing 128Kbit/s of using both channels combined this would transfer the same 29GB of data in about 21 days.

If you needed to transfer this data particularly far. Say from Alaska to Argentina (Practically to the the other side of the world) you would need to drive pretty fast to beat ISDN. One guy did this route as a road trip and documented it taking 23 days.

But if you didn't have ISDN and had to do it with 56K dialup modems it would take at least 3 times as long so it would now take about 1/4 of a year to complete. Dialing directly would probably also cost more than actually driving your truck there.

But today we have tiny SD cards that can hold >100GB so you can still use your truck to race against internet available to mere mortals and win.

[Koen]

but instead have loads of RAM, like for example 256-512 GB of RAM

I left the game a while ago so this quote led me to search where we are now and if anyone's interested : 24TB at SuperMicro servers, 12TB at Amazon EC2, 2TB at OVH rentals.