I've been running a fair amount of infrastructure with them for a while now and I can't possibly recommend them highly enough. I remember there was a time a couple years ago when there was a queue to get a box with Luke. You have to know what you're doing and there are no training wheels attached, but they're awesome and has always been hugely helpful. I figured I'd just start out on them (being cheap) and eventually move off: but networking's been rock solid, and over the past couple years I've been able to work with them to get custom boxes for specific needs. Once I got a SMART condition warning on a dedicated box and they were on it basically instantly.
They know how to manage their network, they're colo'd in multiple top-tier DCs, and the prices are just impossible to beat. I'm able to run infrastructure comprising ~500GB RAM for literally 3% of the cost of EC2, with hugely better disk io to boot.
I'm not trying to dissuade you from using prgmr, but EC2's reserved instance pricing makes it a lot more competitive. You can get the 3.75GB medium instance for $55/mo (1-year commitment including the up-front fee as a monthly cost) which compares nicely against prgmr's $68/mo 4GB instance ($14.67/GB vs $17/GB).
In fact, I'm wondering how you came about your "I'm able to run infrastructure comprising ~500GB RAM for literally 3% of the cost of EC2" calculation at all. Even without reserved instances, Amazon has their 3.75GB instance available for $115/mo (or $30.67/GB). At $17/GB, prgmr is more than half the price (~55%). That is a savings, but certainly nowhere near 3% of the cost.
IIRC, prgmr charges $4 plus $1 per 64MB of RAM. Assuming that they'd sell you a 500GB instance, that would be $8,004/mo. Since you'd probably have to spread it over more machines, it would likely be a bit higher. Anyway, even without dipping into reserve pricing (using the quad XL high-mem instances), you can get 500GB of RAM for $9,000-$10,000. Again, prgmr is a bit of a savings, but in the 10-20% range.
And here's the big point. If you're using AWS for on-demand instances in that range, you're likely to want the hourly billing and rely on the fact that Amazon has much greater scale. Prgmr can have a queue for more capacity so clearly they're not accommodating someone like Netflix who needs 1,000+ boxes to do encoding for a week when a new platform comes out and then shuts them down until a new platform comes out. If you're running a more predictable infrastructure (which it has to be to an extent to go with a smaller player), then you can use Amazon's reserved instances. Amazon's reserved instances beat prgmr on price.
Now, I didn't calculate in storage or bandwidth into Amazon's costs, but even with those additional charges, I don't see how you came upon prgmr being 3% of the cost of EC2.
For my own personal server, I have an EC2 Micro instance with 1-year heavy utilization. That means that I pay $8.77/mo for 613MB RAM - a discount against prgmr's $12. Now, I have to add storage to that - the 12GB that prgmr offers would cost an additional $1.20 - and bandwidth, but I don't break $12 for my usage. If I wanted to commit to three years, the cost would be even less - $6.38/mo.
None of this is to take anything away from Luke - he's literally written "The Book of Xen". That said, Amazon's pricing isn't much more expensive. Their reserved instances can be a good bit cheaper. Amazon also has the scale to accommodate you more easily. Prgmr can't keep thousands of spare boxes on hand and they (as you noted) have had queues for instances. Amazon is by no means the be-all and end-all of Xen instances, but it's pricing isn't that bad (and in fact can be better) and there are advantages to using a company with Amazon's scale.
The comment about SMART errors would indicate they are a dedicated server customer. I don't have many of those, so it's funny I don't remember (and I probably owe them a price cut; I have new and dramatically cheaper dedicated servers now[1] and I try to give old customers first hack at price cuts.) - but the ram/money ratio is a bit more favourable to the customer on dedicated servers (of course, you have to buy a whole lot more.)
It's interesting that the amazon instances are that cheap, though; I hadn't done that math for a while; It does look like I need to improve my ram/dollar ratio; I need to stay cheaper than ec2 for the reasons you stated. (also, ec2 has an awesome reputation and can get corporate customers... so they are going to price themselves rather higher than the cost of production for that reason alone. I should be cheaper than they are.)
Oh yeah, I also do co-location[2] and if you want to talk about saving money? if you need north of 32GiB ram, you can save a huge amount of money over a year by buying hardware and co-locating it. Actually, the hardware cost of a dedicated server? as a rule of thumb, it's less than what you charge the customer for 4 months rent. (Of course, you have to pay for power out of that, so I don't pay off the server in four months, but you get the idea. It's not particularly low-margin.)
I have a couple of the xen instances, but the vast majority of my capacity is on colo'd machines and a managed dedicated box. My total hosting bill is in the $300-$400 range. Admittedly I haven't looked closely at EC2 reserved pricing (I used the on-demand prices to get to the back of the envelope 3% number), but I still don't think it would be close. And this is month-to-month.
they accomplish the same goal; If you are taking the flexibility out of IAAS (please don't say "cloud") through reserved instances, the biggest difference is that you need someone on call to physically handle hardware if you co-lo.
I mean, I agree that they are very different products, but IAAS and a vps are different products, too. (with a VPS, the customer expects me to preserve the data on disk. With IAAS, the customer expects to be able to spin up new nodes on demand.) If anything, co-lo is much more like VPS or dedicated servers, in that you generally use raid and expect the RAID to be fairly reliable.
But if you are running a webapp, just like most webapps that could run on a VPS can run on IAAS and vis-a-vis, most webapps that can run on a dedicated server or IAAS can also run on co-located hardware you run.
The cost difference is, well, dramatic. Of course, the stress difference, if you are notified that you made the front page of hacker news Thursday evening and have to come up with more servers same day on a friday, and your application can't run on IAAS? The stress difference can also be considerable. so I'm not saying colo is always the answer; I'm just saying that if you are looking at hosting options for your webapp, colo deserves a look.
Of course, if you do co-lo, don't forget redundancy. Good hardware you own with RAID is going to be significantly more reliable than a single IAAS instance, all things being equal, but you don't run something on IAAS without redundancy, and if downtime is expensive, you won't run it on hardware without redundancy. For some applications, it's enough to just have spares and carry a pager (of course, that's also a pain in the ass. I better be saving a lot of money to carry a pager) and for other applications? It's okay to just wait until Monday and get the people you bought the computer from to fix it. In those cases, I'd practice spinning up your app on IAAS so you have something, just in case.
They know how to manage their network, they're colo'd in multiple top-tier DCs, and the prices are just impossible to beat. I'm able to run infrastructure comprising ~500GB RAM for literally 3% of the cost of EC2, with hugely better disk io to boot.