I'm okay with the concept of trading cost/time for convenience, and why that might work for some folks, but even before we get to that argument, my experience with Heroku is that it just isn't reliable enough for client sites. Since reviewing it (over the last few months) we've had a couple of instances of real downtime (ie, greater than 30 mins), and a few spots of smaller amounts (a few mins each). We don't get that from our Linode+Puppet sites (taught myself Puppet as we went along, I'm a dev rather than an ops guy really).
Well I guess you are lucky then because Linode has had massive downtimes in their London and Dallas data centres in the last few days and when I was a customer their Fremont data centre went down all the time.
Not sure if the London one was really "massive", I have 12 Linodes in London all on different machines and I had one become unavailable during the downtime a few days ago and it was resolved very quickly.
