What I've had firsthand experience with is:
* Tier-1's profile all their traffic, either directly or with flow export, and will get alerts if they see spikes to certain netblocks, or spikes to specific /32s with specific characteristics.
* Inside a Tier-1, the third-tier support people usually have additional monitoring they can enable for customers if an incident has been escalated to them. They get hundreds of these calls a week.
* A lot of Tier-1's can quickly reroute traffic to a specific /32 to run through special-purpose filtering setups, which might be a DDoS box like a Cisco/Riverhead, or an IPS like TippingPoint, or even just a Cat with lots of TCAM space set aside for filtering.
* A lot of Tier-1's --- maybe all at this point --- have some mechanism set up to share signatures of attacks so they can push filtering further upstream. When I watched that stuff happening, the sense I got of it was that this was really reserved for things like global botnet C&C.
What I'd add to the discussion on products is, most of what's built to combat DDoS is really only useful if you sell transit. If you're a Fortune 500, I know there are ISPs where you can get Cisco or Arbor gear deployed on the head end specially for you. But that's a Fortune 500, not 37 Signals.
I spent some time googling around and in case anyone's interested, these links seem to cover most of the mitigation patterns.
A slideshow covering common mitigation techniques, http://www.slideshare.net/intruguard/10-ddos-mitigation-tech...
A Cisco whitepaper (which, as usual, has en emphasis on Cisco kit and a long URL), http://www.cisco.com/en/US/prod/collateral/vpndevc/ps5879/ps...
A fairly comprehensive PDF document - A Survey of Active and Passive Defence Mechanisms against DDoS Attacks
http://www.fbi.cqu.edu.au/FCWViewer/getFile.do?id=17921