I work at Flickr and I see they mentioned Flickr's ticket server idea, (ab)using MySQL's autoincrement and "REPLACE INTO" trick and mentioned that a con was the write bottleneck.
We're generating more GUIDs than ever with this system and those boxes are more or less idle on every metric. They're right in that we don't meet their time-ordered requirement, but I just wanted to say that writing (or reading) is not a bottleneck.
I like that you guys are using your database's stock features to accomplish this. Most of the time you don't need complicated systems to get things done. Reducing that mental overload of YET another system is huge.
We're generating more GUIDs than ever with this system and those boxes are more or less idle on every metric. They're right in that we don't meet their time-ordered requirement, but I just wanted to say that writing (or reading) is not a bottleneck.