Open Connect | Hacker News Reader

Open Connect

openconnect.netflix.com

47 pointstosh39 comments

Reminds one of the 90s and Akamai

It's a good move for netflix because

- side step net neutrality rules. If it's likely a tier one carrier will screw over your traffic, just store the important stuff with the tier two carriers...

- AWS is pricey what we way you look at it, and network costs can only go up. Amazon is a prime (!) competitor for streaming and it's a bad idea to put all your eggs in your competitors basket

- Disney is coming over the hill. They have great content - i mean insanely great. But I bet Disney sucks rocks at hiring network SREs. They will get there eventually but if you want a huge differentiator between you and the House of Mouse, being the one without a spinning egg timer is pretty good - especially as there is no chance you can match their content.

> and network costs can only go up

I would assume they'll go down. Do you mean because traffic will increase, or do you think cost per byte will increase?

The expensive part of any hosted service is the bandwidth out - and at Netflix levels it must be stupid expensive. Even with the sort of deals they must have cut with Amazon, off loading to local caches is going to reduce someone's costs by a helkuva lot

And yes. Amazon will happily squeeze them till the pips squeak, so their network costs will only go up. (Even if routers etc push the theoretical price downward)

What Disney content are you referring to? In my mind Disney has Star Wars and movies for children. The only great thing that comes to mind is Pixar, but I am not going to pay monthly for two handfuls of movies. But seeing as you are so enthusiastic about their content, it must be my ignorance.

So, again and from a place of interest, what content are you referring to?

Lets assume you've been living under a rock because otherwise your comment is immensely condescending. & star was is for kids.

https://en.wikipedia.org/wiki/List_of_assets_owned_by_The_Wa...

or more specifically movie franchises

https://entertainment.ie/cinema/movie-news/heres-an-almost-c...

I am sorry that you don't like their content, but for what, 95 years Disney has produced decade after decade films that are popular, profitable and proven to be rewatched (and re-paid) decade after decade - films that were made 80 years ago are still bringing in millions - that's what you want for a steaming service.

I didn't say I don't like their content. I am saying that I wouldn't pay every month to have the option to look at their content. Will you be watching Toy Story and Finding Nemo 4 times a month? I'd rather buy those on a disc (which I did) and pop them in every now and then.

Better yet, sometimes cinemas replay those classics so you can enjoy the full experience. I'd rather pay to go look at that.

I'm not convinced the economics work out that well. Rough guess, Disney now has call it 60 top flight movies, from Snow White to Star Wars (top flight as in have been either box office smashes and/or perennial money makers).

If I could buy all the DVDs (I'm not sure I could) I would be paying 15 quid a shot - that's 900 pounds. At 9 quid a month subscription that's 8 and a half years before I break even.

I am fairly sure I lost most of the DVDs I had 9 years ago in our last house move.

Yes it is likely to be bad economics, but it's not that bad.

Edit: I wanted to check my 60 movie count and I think I am under estimating - (1) lists 57 "Classic Animated Movies" - and some are obscure but I would lay good money that most people you know have seen at least 40 - and that does not include any Pixar (20 movies?) Marvel or Star Wars (must be 30 by now).

A family (and yes that's the target audience) could have a Movie Bight once a week for ten years on this subscription.

I guess only time and competition will tell

1) https://reelrundown.com/animation/all-50-disney-classics-mov...

A lot of Marvel content I would presume.

Which is also not great.

I mean, you might not like them but they are among the highest grossing movies of all time.

I get what you are saying, but I'd find it stupid to pay a set amount per month to see the same movies over and over. Do you really think people will pay to see all of the Avengers movies every month? If yes, then fine, good for Disney. I just find the 'they have better content' line a bit much.

I'd rather watch Master of None, Fargo, The Crown and many more quality Netflix than some action here movie that has been cut from the same comic book cloth as the previous one. But that is just a matter of preferences, but to say their content is better based on how good they do in the box office is nonsense.

I think we can all agree here that this segregation of content feels like a step back, and the grandparent asserting that Disney has great content that warrants paying for is controversial in this light.

It’s besides the point OP wanted to make.

It would be nice/interesting to have a Netflix appliance locally in the home network which cached content that was expected to be viewed (next episode of a series, suggestions, promoted shows) along with neighborhood peering to keep as much of the traffic along the edge as possible (i'm not sure where ISPs measure bandwidth if you are connected to another node connected to the same local hub).

OCA seems like a great model for static content, but I wonder if it supports live streaming content as well. When I spoke with Limelight several years ago, they had a reasonably nice model for fanning out streams hierarchically so a single feed could be fanned out very widely late at the last mile.

Upstream traffic from the home (especially in DOCSIS) causes more congestion than downstream traffic from a CDN so CDNs really are better than P2P today.

What is it about the upstream that causes more congestion? And as you say "especially in DOCSIS"?

I'm not aware of anything with DOCSIS that requires the up-speed to be so low. It's just a matter of the cable ISPs not using as many channels for up-speed because on average a residential connection doesn't NEED a ton of up-speed.

Now if they're just not building out their networks to have a lot of up-speed bandwidth at the node level then that's another discussion.

It really seems to me that much of this content is likely consumed by many households within a node. It does seem kinda foolish that it's streamed multiple times to multiple residences within a neighborhood when the content could easily live somewhere on that node. I guess it just comes down to what is cheaper? Bandwidth or storage? Seems they're deciding it's bandwidth which makes one question why we need data caps on residential broadband.

Due to the frequency split there are fewer upstream channels available and changing that would require replacing all the filters. Downstream bandwidth inside the ISP is definitely cheaper than anything.

But my point still stands. That was a design decision. There is nothing technically stopping them from changing it if they felt they could ease the strain on their downstream bandwidth by using more localized caching and even P2P schemes.

In the end it just further highlights that downstream bandwidth is cheap if they don’t mind sending 20 copies of Stranger Things in 4K down the pipe to a single node covering a neighborhood. So why the data caps?

>When I spoke with Limelight several years ago, they had a reasonably nice model for fanning out streams hierarchically so a single feed could be fanned out very widely late at the last mile.

So... multicast?

One of my back burner ideas is a “home WiFi router with apps”, and this type of caching/peering is precisely what I’d hope to enable.

I started building something like this at one point and called it AppBox [0], to make installing apps on the device similar to installing apps on a server. I think Docker took over this role.

[0] http://appbox.rkeene.org/

This is probably one of the reasons why Netflix is so much more stable and reliable than pretty much every other streaming service that I've used more than a couple of times.

Oddly enough this strategy has been done before. Akamai did the same thing back in the day. ISPs were getting hammered with traffic so Akamai sold edge caches to the ISPs then also sold the CDN to websites and what passed as a “content provider” back then. Still super cool.

I had the privilege of partnering with Sanmina a few years ago, and they had the most innovative designs I've seen to date for storage.

My favorite one was where you could stuff equivalent of 30 2.5" SSDs in 1U.

I find it strange that this page is publicly available–wouldn't Netflix be the one reaching out to IXPs to get their content cached?

As someone at a small ISP, this being public was super helpful for us. Caching Netflix is (I think) better for us than it is for Netflix. We saved a ton of bandwidth during peak hours by getting an OCA in our network.

We are too small for Netflix to ever reach out to us, so this was perfect.

The Open Connect program is maybe 10 years old. Is there anything new here?

I remember Comcast whining about their transit traffic getting saturated by Netflix during the first big Net Neutrality fight. Open Connect solved their problems for the price of electricity to run the box. That wouldn't allow Comcast to hold their customers hostage from Netflix though.

I suspect that if netflix went to the ISPs to cache their data, the ISPs would be more than happy to charge them for it like to do with other content networks. In this case, netflix is assuming that the volume of netflix traffic at the ISP will drive the ISP to request on network caching. Netflix is happy to provide this, but is not willing to pay the ISP for the privilege.

Does netflix care about how much traffic they are directly serving vs how much is hitting caching servers? Are they playing chicken with the ISPs hoping the ISPs will blink first and put caching servers servers onsite for free?

This is most likely a win-win senario for both parties.

Netflix can be more stable, and can save on bandwidth in their own DCs.

ISP can save a lot of egress bandwidth by having Netflix inside their infrastructure.

I would be surprised if any of the parties payed. It's a great strategy to have the ISPs on board with this

Exactly. It's almost like a peering relationship between providers but in this instance netflix is not directly connected to the ISP so wouldn't be paying the ISP for bandwidth used by their customers accessing netflix. It definitely saves the ISP a lot of bandwidth cost

This is spot on. This is basically how Netflix gets free colocation. Netflix's monopoly is what allows them to do this. ISPs would laugh if anyone else wanted free hosting of their edge equipment, but because Netflix is such a burden on ISPs, they're able to strongarm ISPs into providing free hosting of Open Connect appliances.

It's not unusual to do this. Cloudflare has an open peering interface: https://www.cloudflare.com/partners/peering-portal/

I too find it surprising but only for competitive reasons: do you want Disney/Hulu and HBO etc to know how you do it?

It's not really something netflix even came up with. I know I saw Akamai boxes in ISP datacenters way before netflix got into streaming

Most CDN operators do this, including apple which also has a public website about it - https://cache.edge.apple

That is true. I didn't think that they were doing it was the trade secret but the way they were doing it might be, unless that is pretty standardized too. I don't really know.

The fact that they have and deploy boxes isn't the secret. It's the hardware configuration and software integration which is the sauce.

Can someone ELI5 this to me? Netflix is providing devices that are designed to route Netflix traffic or data only through an ISP's network, reducing the strain on the entire ISP's service?

It’s storage cache they put in your isp so if you and your neighbor watch the same thing, it only goes the first hop from your house to the isp instead of all the way to Netflix, which is much cheaper for the isp. This is a fairly common thing that cdns do

Reply on news.ycombinator.com