As a veteran of such comparisons, one thing they can never capture very well is what happens when the “shit hits the fan”. It’s easy enough to compare on price and features, but what happens when something goes wrong at 3AM? How quickly to you get responses to emails? That kind of thing is more difficult to quantify, but very, very important.
I think this is more or less the conventional wisdom: on a machine-by-machine basis where you’re treating a machine as a VPS, AWS isn’t really competitive on price/performance (neither are the other big clouds, like Azure, Joyent, or Google Cloud). You use AWS-style clouds if you need something they offer, like their much more extensive tooling, ability to flex capacity, etc., or if your organization is already using them for other things and you want to consolidate. If you just want always-on standalone servers, the VPS providers are a better deal.
Exactly. Amazon is not cost-effective on a per-server basis. You use it because you need to go from two servers to two hundred servers in five minutes, and then back to two again in an hour. (Or you waste a ton of money, like my current employer.)
Unfortunately, even Amazon and their own tooling seem not entirely on board with this, with the focus on “instances” (where their service is least competitive).
One thing that’s a bit annoying as small fry is that you seem to have to be big to do it that extensively. There’s a rhetoric around The Cloud that now massive computing power is totally accessible: “anyone” can get 100 servers for 1 hour for like $2, or 200 powerful servers for $20, making once-rarified compute clusters within everyone’s reach. But by default there are actually quite low instance limits, and they will only raise them significantly (to 100+ range) if you’re spending significant money. If you’re a researcher who sometimes likes to have 100 servers for 1 hour, they don’t really like to do that.
I can understand that as a business decision, because it introduces a bunch of churn and instability in their offering for small amounts of money. So they reserve ability to do a lot of flexing to customers who are paying them enough to be worth the ops hassle. But we’re not quite at true “utility computing” yet, where anyone can access arbitrary compute just by paying pennies-per-instance.
Imagine if there were a blog post like “Crack the NSA’s RSA key for $20” with a script you just plugged your AWS creds into. On the one hand, kind of cool. On the other, prime example of the instability you mentioned. Amazon probably doesn’t want to keep so much spare capacity around that they could handle a thousand strangers spinning up a thousand instances each for one hour. Tomorrow, after everyone’s had their fun, that’s literally a million servers sitting idle.
It’s probably worth comparing to dedicated hardware as well, for the standalone server aspect.
For an initial outlay of about $200, I picked up a server that exceeds digitalocean’s $160 month plan, but I only pay $60 a month hosting. There’s obviously more lead time required for an upgrade, but the $1000 saved in one year can buy quite a capable upgrade, and my monthly costs after that remain the same.
I used Linode’s $20 plan for a while because it’s hard to find physical hosting that cheap, but there’s a big elbow after once you move beyond that.
Exactly. Linode is great for 1GB to 2GB DDR3 at 10 to 20$/mo, but when you can get so many powerful month-to-month dedicated servers elsewhere, I’m not sure who would go for Linode’s higher-level plans, which don’t even offer any discount (in terms of bang per buck) compared to their 10$/mo offering, which is arguably thus the best deal, considering you get their support and unlimited DNS manager (which works great for the slave zones).
And I think co-location and dedicated servers, too, require a comparison — it would appear that oftentimes it’s actually cheaper to rent a whole dedicated server than to buy your own for colocation. E.g. online.net servers in Paris currently start at 5,99 EUR/mo, 15,99 EUR/mo for 8GB DDR3 and 1x 120GB SSD on a Supermicro server, or 29,99 EUR/mo can even get you a Dell-branded server with 16GB of RAM and 2x 1TB discs with only a minor setup fee of 40 EUR (and a month-to-month contract), WholesaleInternet servers in Kansas City go for 24,00 USD and up with no contract and no setup fees, whereas colocation deals can rarely be had for less than 50$/mo.
I like dedi’s so much that I’ve even created a list of these cool providers — http://dedi.su/. And, btw, prices at hetzner, online, ovh etc are so low they don’t even offer any affiliate programmes! Only Linode can afford to have an affiliate programme.
I was looking into self-hosting a server recently (i.e. “colocating” in a house), and it’s surprisingly hard to beat OVH/online.net’s prices even doing that. Just the power for one of their $50/mo servers would cost me a significant fraction of that unless I keep it always idle.
That’s quite the server. :) Around here, that’d be about 7 amps continuous draw. 1 amp is pretty reasonable for a 1U server. Of course, one of the things I’m paying for with offsite hosting. Is the power.
Between an upgrade to “business class” cable and the power (and cooling), home hosting works out about the same. I haven’t done that since college dorm days. :)
Is that 7A @110V or @220V? At 220V, that’d cost me over €300/mo! That’s roughly 1.5 kW, and 1.5 kW · 720 h · €0.30/kWh = ~€330. Admittedly Denmark has quite high retail electric prices (industrial users pay closer to €0.10/kWh).
yeah, 120v. I was also assuming a price closer to $.10. Should have included numbers for clarity.