I own a small ISP in SEA, Bandwidth from ggc is about 70% of our traffic (mostly youtube). While we don't have GGC ourselves yet, our upstream does and we have peering with them. Its amazing, how much of the actual "Internet" goes through GGC. We also have akamai peering, which handles about 10% of our traffic.

Are you at SIX? I'd guess there's at least one of these there, if Google themselves aren't connected. I'd wager Akamai and Netflix are at SIX too.

whats SIX?

We are at SEA (South East Asia) :)

GGC's are everywhere, its relatively easy to get, other than proper business paperworks the onle requirement is to have decent bw, I think minimum 1gbps Google BW.

Similar to what other providers of original content do, e.g. Netflix.

I wonder, if local cache setups would have different impact on jurisdiction, taxes and such. Google is known to always say you deal with Ireland, e.g. outside of the US but what if the cache is local?

Does Netflix have caching beyond AWS data centers?

Yes:

https://openconnect.netflix.com/peeringLocations/

This was somewhat famously part of the network neutrality debate because certain large ISPs were trying to claim that their customers were transferring too much from Netflix and thus needed to be double-billed but those same ISPs refused to install the free OpenConnect servers which would have reduced that traffic.

My understanding though is that if you have an open connect box on your network it can act as a DDOS, stressing all your infrastructure closer to the customers. Netflix eats all bandwidth.

That's not a DDOS – it's called “actually delivering the service your customers paid for”. If you've underprovisioned your network, that might be a problem but it's not an attack.

It's also something of a best-case scenario because the only party affected is the same party which has the ability to fix the problem: since it's on your network, you can add additional capacity or institute some sort of fair-queuing without the need to negotiate with outside companies.

I imagine the Open Connect box would do automatic bitrate tuning for all users on that network for immediate mitigation and notify obviously Netflix and the open connect ISP for long term capacity management.

Oh yeah https://openconnect.netflix.com/

To put a finer point on the OpenConnect comments - Netflix doesn't stream any video from AWS. All video bits are served by the OpenConnect CDN.

Anyone know if Google will allow these to be installed at IXPs? $work (ISP) likely isn't large enough to warrant having one of these on our network but a bunch of us smaller ISPs all peer together at a private IXP (in Indianapolis) and our collective traffic levels almost certainly justify it.

Then again, with tens of gigabits to Chicago (where Google has a major datacenter) it probably isn't really necessary.

(Tedious disclaimer: my opinion, not my employer's. Not representing anybody else. I'm an SRE at Google, and my team is implicated in this infrastructure.)

This stuff is designed to be able to go almost anywhere that's willing to host it under the standard contract. File a request, let the relevant team figure out if it's worth putting a rack in. Nothing bad can happen as a result of asking the question.

(Ah, "major datacenter" in Chicago. It's bad of me to be amused by this. We don't have anything we'd call a "datacenter" in Chicago, that's a small installation ;). The list is here: http://www.google.com/about/datacenters/inside/locations/)

Yeah, sorry, I know there isn't a "real" Google datacenter there (too late to "s/datacenter/presence/" on my original comment now). I don't technically have any of my own datacenters, but I'm present in a few, just like you guys have a "major" (to me) presence.

I've put them in a few ISP's over the years, they do support multiple ASN's to be served from a single cluster so don't see any technology limitation there.

The traffic requirements were around 2Gbps of peak Google traffic last time I checked with them, so you might actually qualify already, certainly not as restrictive as the Netflix caching servers where they want 10Gbps of peak traffic.

Think your main problem is that Chicago is so close...

In YouTube if you right-click on video you get 'Stats for nerds'. Does anyone know how to translate the stream hostname to actual IP / geolocation?

${hostname}.googlevideo.com

$ dig r1---sn-4g57knez.googlevideo.com +short r1.sn-4g57knez.googlevideo.com. 173.194.6.166

The videos are split among a couple of domains/hosts. uMatrix gives a nice overview over the used domains.

Is this what AppEngine apps / Google cloud storage files are served from, or is that a different CDN?

That data would be served from Google's own datacenters.

To be as efficient as possible, these devices would only serve content that is accessed by many users (e.g.: popular Youtube videos).

From https://peering.google.com/iwantggc:

"GGC is implemented as a set of servers deployed in your datacenter, remotely managed by Google."

Basically you're leading the greedy little boy Google by the hand to the back door of your house, unlocking the door, and going on about your business. "Don't you go in now, kid".

Feeding the troll, but... many (most?) GGC nodes go in ISP networks, where there are already plenty of other third-party systems (Akamai, Netflix OCA, etc.) and where the traffic is all "hostile" anyway since it's literally the Internet.

Meh, I work for an ISP. We have lots of hosts in our facilities / on our networks that don't belong to us. If we don't manage it ourselves (i.e., if it's not one of our devices / servers), it's considered untrusted -- that includes customer connections, co-located servers, and systems like these. From the ISP point of view, these Google servers are no different than anything else.

You have to be a bit more specific. What risks do you see that sending traffic to google via the internet doesn't have?

Then why "Google Global Cache is available by invitation only."?

I don't know who down-voted you and why.

This other product has been around since many years ago (pre Snowden), and people has been connecting it to intranets, BI systems, etc:

    https://www.google.es/intx/en/work/search/products/gsa.html

More on the same to think about what you commented.

It's just evidence-free conspiracy babble and you're making it worse by mentioning Snowden in a misleading attempt to imply they're working with rather than, as Snowden showed, being attacked by the NSA.

More broadly, however, these things are cache nodes – putting one in a data center makes sense but that's not saying you need to give it the privileged access stereokai implied. The win comes from deploying it on the same physical network but you'd still want to firewall it like public internet traffic.

Handing your internal data to Google appliances to process vs redirecting traffic you'd otherwise send to google to their appliances is a "slight" difference, don't you think?

I guess google could be putting bombs in them, but somehow I doubt it...