News.com's Martin Lamonica has written a two page report on Amazon's utility computing initiatives under the title Amazon: Utility computing broker. As many of you know (because of the disclosures I often print at the bottom of these blog posts), I'm part owner of an event production outfit called Mass Events Labs. Under that name, my partner Doug Gold and I produce events like Mashup Camp, Mashup University, and Startup Camp. By the way, Mashup University and Mashup Camp 3 will take place at MIT in January. Due to facility limitations, there will only be room for 250 participants. Please see the Mashup Camp Web site for more details.
The reason I'm bringing this up here is that, for all intents and purposes, Mass Events Labs is a business just like any other business that has IT needs and when, as a business manager, I do the math on Amazon's utility offerings, it's pretty much impossible for me not to consider trying them out.
Given how integral our Web sites are to our events (attendees update the Web site during the event), I decided that we should be careful about taking any risks with our system hosting (Web, email, wikis, blogs, etc.). I have seen or personally experienced enough problems with hosting companies (particularly the low-end ones) that I felt in my heart that I'd rather be safe than sorry.
Today, at a cost of about $700 per month, Mass Events Labs has two Dell servers dedicated to running various Web sites. The servers themselves have all sorts of fault tolerant technology built into them (redundant power supplies, fans, network interface cards, etc.). But to be extra sure, we have load balancers in front of both servers and we keep the directories and databases between the two in synch so that if one server dies on us, the other can take over. The servers are located in a datacenter that, beyond the local electric utility, has not one but two layers of backup power generation as well as redundant high-bandwidth connections to the Internet. If something goes wrong with a server, it gets a human's attention -- not mine (that's peace of mind) -- immediately.
But $700 per month is a lot of money for us. Not that we mind spending it for the peace of mind we get (and the dedicated servers, which is important to us for other reasons). But, as a business, it is our largest monthly recurring cost. So, when I first heard Amazon CEO Jeff Bezos talking about Amazon's EC2 elastic computing initiative ("EC2" stands for Elastic Compute Cloud) at MIT's Emerging Technologies conference, I was listening not just as a journalist, but as a business manager too. For 10 cents per hour that you'd be running an "EC2-based server," Amazon offers the equivalent of a 1.7Ghz x86 processor-based system, 1.75GB of RAM, 160GB of local disk, and 250Mb/s of network bandwidth. Word has it that Amazon is using Xen under the hood to offer these virtual servers to EC2 customers. But more importantly, the 10 cents per hour thing is worth plugging into a spreadsheet for just about any business -- Mass Events Labs included -- that thinks it already has a pretty decent hosting arrangement. So, let's do a little math.
There are 365 days in the year and 24 hours per day which means that there's a total of 8760 hours per year. Assuming that your server has to run all 8760 hours per year, the total cost per EC2 server per year is $876. A real analysis of EC2 versus other hosting options isn't quite that simple. If you read Amazon's fine print, you'll may end up incurring some storage and bandwidth charges as well (the fees are similarly nominal) and you may have to spend time with someone from Amazon to understand the extent to which EC2 server instances, once the program is out of beta, will be fault tolerant.
EC2 leverages the same infrastructure that Amazon.com has built for itself. It involves multiple datacenters, network and power redundancy -- all the stuff you'd expect from a commercial grade hosting outfit. Depending on my comfort level with EC2's reliability (I haven't even figured that out yet), we may not even need a failover server as we have now (or, in EC2's case, that would be a failover "x86 instance"). But just assuming we were comparing apples to apples (as best we can) and we assumed that Mass Events Labs (or your business) needs two servers, the total annual cost is $1752. Today, our annual cost for two servers is $8400.
And Mass Events Labs is a small business. What about you bigger ones out there? The ones running 50, 100, or a 1000 servers. Suddenly, if you're currently in the commercial-grade hosting business, things are looking pretty bleak with this new kid on the block. Let's say you have 100 servers that you run around the clock with a commercial grade hoster, the annual savings moves into the hundreds of thousands of dollars. What? You have 1000 or more servers? Now we're talking about saving millions.
One other side note: none of these back of the envelope calculations take into account what happens if you get smart about server utilization and decide to take full advantage of the Amazon APIs that, in a blink of an eye, can turn these x86 instances on and off. With dedicated hosting of the sort that we have, because of our annual contract, we're married to two servers for an entire year. Whether we're using them or not, we're paying. Not so with Amazon's EC2. With EC2, you can use Amazon's APIs to programmatically scale your server horsepower to match the need at any given point. This is where some savings can really start to kick in. Maybe those 10, 100, or 1000 servers don't need to be running 365x24. Maybe, a bunch of them can be in the "off position" more than they're in the on. In which case, you don't pay the .10 per hour for them. Model this into your spreadsheet and watch that number on the bottom line get smaller, and smaller, and smaller.
I'm not saying that EC2 is for everyone. But, compared to the cost of traditional server outsourcing (let alone running your own servers on your premises), how can you not be opening your spreadsheet software right now?