I'd like to migrate or even perhaps add a secondary DNS but RS DNS doesn't seem to even offer zone transfers (the best you can do, I guess, is to use the API to get your records out).
Its tough to do. DNS is a dumb and ancient protocol from the "lets all be friends" days of the internet. Its a bit more complex than installing mod-evasive and calling it a day.
DDOS is currently an unsolved problem for many popular protocols and services. Blaming just Rackspace seems unfair. Last week in was Namecheap. The week before it was someone else, etc. DNS DDOS is just a non-trivial problem to solve.
This should also be a reminder to have more than one DNS provider in your domain record. A backup nameserver from a different provider saves you and your customers a lot of heartache.
I'm not up to date on what that number is these days, but a few years ago 100Gbit/s was the target. For an Anycast DNS service, you don't need that in every site -- if you're willing to take small sites offline and degrade your service's end-user performance in this scenario. You will still need it in 2-3 sites, probably located in US west / US east / EU west because of the cost of datacenter space and the availability of cheap connectivity.
Serving 100Gbit/s requires 20 10G circuits in each site (~$15-30k/mo), 80 servers (2-3 racks, so $5-10k/mo), and a not-so-cheap router.
So your monthly opex is $20-40k per site, $60-120k worldwide. Your capex is $450k per site, $1.35mm worldwide.
It's not an insane amount of money, but it is kind of a lot to spend on DNS. It's particularly tough to justify at a lot of companies that you need this kind of buildout, when normally you can operate on 1/20th that amount of hardware.
The real point I want to make is that the only companies really putting this amount of money into DNS infrastructure are DNS providers themselves. I know of no hosting provider/ISP/etc. who puts nearly this much thought or effort into DNS.
Even for myself - and I consider myself a DNS junkie - my anycast cluster was a dozen or so servers worldwide co-located w/ a mid-market CDN provider (plus a few locations from my own ASN). That setup was greater than 99% of DNS installs out there - but it would have fallen over under a major attack. It truly takes some skill and financial commitment to build out correctly.
1. DNS hosting companies (Dyn, Amazon, etc.) 2. Large Internet companies (Yahoo, Google, etc.)
And for sure not all of them! A former employer's public zone was hosted by a DNS hosting company that didn't use Anycast, didn't have a footprint outside US/EU, and seemed to have little to no idea when or why traffic spikes happened.
So to do DNS really right you need a fair amount of money, some specialized expertise and knowledge of some non-DNSy things (Anycast, among others). I'm really not interested in hosting my own (public) DNS these days at work, it's just not worth the time and effort until you get gigantic. As with e-mail, there are specialists who can do it better than (most) places can.
I do kind of worry about depending on one DNS provider, so I've tried to use two when possible. That's a whole other world of hurt though.
So for example (very high-level) you might have a server in the US, and one in Europe. If your are eating a DDoS attack maybe the botnet has tons of US-based IPs, but very few euro. In that scenario, you will have all your US customers down but the EU guys are just fine.
That's pretty simplistic though. It's less by country, more by ASN. And it's also a case of automation killing you. What happens when the DDoS attack takes out a single PoP worth of DNS servers? Usually you have some automated tooling that will take down that PoP entirely, and withdraw the announcement. You want this behaviour in almost all cases. However, now that attack traffic just got sent to the next-closest server (in this example EU - taking down the whole worldwide cluster).
So it's not as simple as having "100gbps worth of inbound DNS capacity" - you need to spread it out properly, and of course deal with attacks like you would otherwise.
The other problem is it's UDP - so spoofed packets and the like are a much bigger problem.
It's a tough call to say if it's better to protect the rest of the network by trying to isolate the attack, even though that means you are totally down in one region. I'm inclined to keep 1/2 running than to have nothing, but if I have this release valve of moving traffic to the farms designed to soak up abuse traffic then it makes my decision easier. I should be able to get back up worldwide, if I've done my provisioning properly.
One of the nice things about this kind of attack is that your attacker has no choice but to follow your routing, whereas if you use your DNS GSLB to re-route an attacker away from your HTTP server farm (to handle a different sort of attack), they may not stop hammering on its IP address. After all, it's pretty unlikely that they respect DNS TTLs.
This approach is simplistic, but it's basically the only solution. The nature of a DDOS attack is that you cannot predict its origin, beyond the distribution of compromised client machines around the world. E.g. your attackers are pretty unlikely to come en masse from Africa.
Spoofing is an issue. Not much to do about that at your farm level, but you can start working with your upstreams to get ACLs in place to prevent some portion of that. And if the attack is prolonged and sufficiently large, the only way you're going to stop it is to identify sources and get help from involved networks and ISPs.
Market cap and technical sophistication, esp at large scale, tend to become divergent. In other words, it is generally assumed that big companies do things poorly. That's the entire notion of startups.
I will say this: Dyn can probably do a better job of hosting your DNS than Rackspace, Rackspace is not focused on DNS, and I'm not aware that they even really do anything special except for running lots of big DNS servers. Even when I worked at Rackspace 10y ago, the size of their customer base made them a constant target for DoS attacks. Do you need Dyn? Also probably not.
Rackspace has patented DDos prevention technology called RiverGuard, but since being patented only illustrates that it is novel, and not that it is of any quality, that is also meaningless.
I will say that, even though I prefer as a past employee to work with other companies, because the things Rackspace is bad at are things they don't give a fuck about solving, they run a damned fine network.
Further, as a replacement for / alternative to a colo, it is irrelevant whether a company can run a good DNS infrastructure, because that's not what I'm asking them to do. Rackspace DNS is there for convenience.
Cloud providers like AWS use their own cloud infrastructure (we think) to host services like Route53, which are also not infallible, but tend not to _entirely_ go down like this. Rackspace, while being somewhat competetive in the cloud space, doesn't operate anything like Amazon, doesn't do anything in a very distributed manner, and is backing those cloud offerings with what I expect to be vertically scaled DNS servers on top-end Dell hardware behind some kind of load balancers.
Money doesn't make you good at things, rather it helps you to not have to worry about the details, so it is inversely true that a company with a substantial market share should be expected not to make the same mistakes as the little guys. They're operating all day, all week, all year under the fallacy that their scale is proof of the quality of their work, and following this logic, they don't have to do any work at all. ;)
All we know is that Rackspace has dns servers in three out of its six datacenters[2], and that they appear not to have always-on DDoS protection in place.
Market cap and technical sophistication, esp at large scale, tend to become divergent.
You say this, but go on to compare Rackspace unfavorably to Amazon who, at 141.93B (of which AWS is just a part), has over 20x Rackspace's market cap and probably 10x the infrastructure[2][3]. Big companies do things poorly, but then it's hard to have highly-available dns without having lots of infrastructure.
Namecheap, Route53, Dyn, DNSimple, and now Rackspace have suffered DDoS attacks with outages of varying severity. Don't act like DDoS prevention is a solved problem and claim technical incompetence.
1: http://www.thewhir.com/web-hosting-news/web-host-rackspace-o...
Yeah, no zone transfers, but you can always grab a complete Bind 9 export via the API or just call support and they can fetch it for you.