Datapath.io: Provider-Independent Elastic IP Addresses | Hacker News Reader

Datapath.io: Provider-Independent Elastic IP Addresses

datapath.io

45 pointssspies22 comments

I'd suggest you scrap your "fancy" website and replace it with two paragraphs of text that explain what you are trying to sell and how it works.

The "vague bullshit in between cute animations" approach really doesn't work at all for this type of product.

When I first saw the page (opened along with a bunch of other tabs from HN links), with its "Timeout error!" balloon and "Service disruption" headline, I thought it was a fancy error page and hit reload. Oops.

Their twitter feed has a diagram describing the process on AWS: https://twitter.com/datapath_io/status/560027255188250625/ph...

EDIT another: https://twitter.com/datapath_io/status/559741202342625280/ph...

This is a clever architecture; I never thought about using VPC direct connect to get out to the Internet.

Doesn't every CDN already accomplish the same functionality with multiple origins and failover detection, while accelerating static content by caching it at the edge? Given how cost competitive CDNs already are, I don't see how they can charge a low enough cost in comparison to be compelling, while still incurring bandwidth costs and at least some network infrastructure which make it almost as expensive to run, minus some caching nodes. And if for whatever reason, there does happen to be some use case for anycasting the IP that's advantageous, any CDN could add this feature overnight while having a much more established footprint and infrastructure in place.

> If your application goes down, all you want is to get it back up. What if your cloud provider is down? Use DATAPATH.IO to route your users to a healthy location while keeping your IP addresses.

But... what happens if datapath.io goes down?

I think the idea is that datapath only functions when you go down. ie: if only you go down, datapath will fix it via BGP magic. If datapath is down and you are up, stuff continues to work.

https://www.youtube.com/watch?x-yt-ts=1422411861&feature=pla...

I assume this is BGP routing as a service (which sounds cool).

But, would this not break the internet if everyone used it? routing individual IPs (rather than large blocks) sounds like it would end in chaos?

In general, most ISPs will drop any routes more specific than a /24. Presumably datapath has negotiated with some specific providers to allow /32s; it's up to that ISP to scale their routers to handle it; other ISPs will only see the /24 aggregate.

> What if your cloud provider is down? Use DATAPATH.IO to route your users to a healthy location while keeping your IP addresses.

Call me crazy, but I'm not sure I really care about keeping my IPs. Route 53 seems as close to bulletproof as we're going to find with DNS, and it's got cloud-independent failover routing built in. Point the primary record at your nice little EC2 server, and have the secondary record pointing as Azure or Google Cloud or your mom's laptop, whatever, and it'll do what Datapath is doing without futzing with your routing. ...right?

You cannot do graceful failovers with Route 53. This is, because DNS is a system of many dependend caching layers and the TTL value is not realiable.

the TTL value is not realiable.

How is the TTL not reliable?

In my personal experience DNS cut-overs have never been a problem. Is there any major OS or ISP doing DNS wrong?

We have seen providers set the TTL value to 300 seconds no matter what. Chrome has been caching names forever while running...just two examples

Do you have to pay for your traffic twice?

... and does this add hops?

Id adds one hop (our appliance) to the path. It tries to increase performance of your path by taking non-standard BGP metrics into account: congestion and latency

Does that mean all traffic passes through your appliance?

Where is it physically located, what about congestion of the appliance itself, what are your peerings?

Traffic passes the appliance at the edge of the hosting provider. It takes congestion at all links and the appliance itself into account and re-routes accordingly. Depending on the characteristics of the location, at least three very unsimilar transit connections are used.

at least three very unsimilar transit connections are used.

What does transit mean in this context; like between my VPC and your appliance? Can you give an example for e.g. an AWS region?

How is the appliance implemented, is it physical hardware, EC2 instances? Is it redundant, how do you scale it in terms of bandwidth?

Transit providers sell access to the whole internet to us. We are in negotitations with them, but we will be more specific, once we have the commitments. Technically, we choose at least three different providers, so our appliance has a good basis of decision-making. For the connection between your VPC and our appliance, we use the regular AWS DC API.

Nope

Thats very cool

Reply on news.ycombinator.com