Second Thoughts

I am re-thinking my comparison of Amazon EC2 versus low-cost dedicated servers. Many hosting providers offer $99 servers with 80 GB of storage space and 1000 GB of monthly bandwidth. I wrote last week that the same combination would cost $284 at Amazon. But my calculation was off.

The $284 price tag assumes that (a) you need always-on processing power, and (b) you are fully utilizing your disk space and bandwidth allocations. Relatively few server users meet these criteria.

For instance, if you're a developer working on a not-yet-live project, your resource usage would be minimal. More interestingly, you wouldn't need 24/7 CPU access. You could stop paying for your Amazon EC2 server instance whenever you're away from your desk. At the end of the month, you might get a $30 hosting bill - for 250 Mbps of data transfer capacity, dedicated processing power and fault-tolerant data storage.

Once your site is up and running, scalability becomes a major concern. You could lease multiple dedicated servers to anticipate traffic spikes; some data centers charge as much as $1 per GB for bandwidth usage beyond your monthly allocation. Or you could deploy additional EC2 server instances as needed and pay as you go for bandwidth.

By unbundling processing power, storage space and data transfer capacity, Amazon gives you full control over every dimension of your computing requirements, instead of forcing you to construct solutions out of inflexible building blocks. And since bandwidth usage is billed a la carte rather than in over-sold bundles, you're not stuck paying for capacity that you might not use.

So I was wrong about Amazon's competitive position relative to discount dedicated servers. It's a viable contender for many customers' business.

Of course, one group that EC2 would not appeal to is high traffic users with predictable bandwidth requirements. Why pay Amazon $0.20 when a $99 server offers $0.10/GB bandwidth - storage space and processing power included?

I wonder how much longer deals like this will last though. If pay-per-use catches on among low-traffic users, how many hosting providers will be able to sell less-than-half-price bandwidth compared with Amazon without these customers' unintended subsidy?

Comments

[dan]

Isabel,

I think every hosting provider in the world has at one time or another said "Hey, why don't we charge based strictly on usage?"

The tech here is really cool... but without flat-fee pricing I don't see it being used to much.

[ScottY]

I don't think Amazon's EC2 is catered for the majority of current dedicated server market at all, i.e. those who grown out of shared hosting service, or resellers looking at starting a web hosting business.

And EC2 is appropriately named -- Elastic Compute Cloud. That's the market they are targeting. Something elastic, where you can instantiate or dismantle at will. Something that computes, which requires lots of CPU resource (and not much else). It's a cloud/cluster, which provides reliability and redundancy. Great for massive computational projects that has fixed life-time, but that's about it (at this stage, at least).

The very fact that you don't even get a static IP address (everything DHCP), and you don't get any persistence other than S3 -- shows it is not competing with today's "web hosting" dedicated server market.

Of course these can be fixed. Amazon can become the world's biggest general-purpose Xen-hosting provider. But we'll talk about it when that day comes :)

[Isabel Wang]

Hi Scott, I like your write-up much better than mine! And I agree that EC2 doesn't seem like a great platform for running a shared hosting business on. I'm guessing Amazon doesn't see "traditional" web hosting resellers as its target market; the question is, should they?

Shared hosting has changed a lot over the past year or so. For one, Microsoft and Google have gotten into the game. I think they'll make a big impact not necessarily in terms of the customers they'll take away from current providers, but by changing end users' expectations on what web hosting should be like. For instance, neither requires customers to install applications, FTP files, etc.

I do see developers building web-presence-enabling services on EC2 - if you think about it, Flickr and MySpace and Salesforce.com all provide "shared hosting", in a way. They give end users bandwidth/storage space on shared infrastructure for putting data online. But instead of 3000 GB/month, they advertise the functionality of their front-end applications, and the non-technical public seems to like that.

So EC2 wouldn't work for starting a web hosting business on, but does an aspiring entrepreneur have to become a web hosting reseller? Or maybe he could come up with a new twist on getting end users online?

[ScottY]

Yup. I agree with you.

The web designer shops or resellers are not going to be attracted by EC2's offering. They need something that can host hundreds of websites for the local fish 'n' chips shops. In some sense a clustered super-reseller package like Mosso might be more suitable.

But that's all so Web 1.0 :) And I don't think that's an area an "aspiring entrepreneur" should be getting into.

So can we conclude that EC2 is a great scalable platform to build/deploy web *applications* on, than to host web sites on?

[Isabel Wang]

Yes! :)

BTW, maybe the local fish 'n' chips shop doesn't need a Web 1.0 site. Instead, maybe it'd be better served by Google Local, where it can set up a Google Maps presence to show consumers where it's located, and offer Google Coupons to encourage visits. With the new Google Apps for Domains service, it can even set up web pages, email addresses, under its own URL. Google doesn't mention anything about traditional web hosting features like bandwidth/disk space or PHP/MySQL/etc, but I doubt the average restaurant owner understands any of that anyway.

[Nemo Semret]

EC2 is indeed more appropriate for people who need scaleable CPU usage on demand. For content, i.e. where bandwidth is a larger cost than CPU, and where it has to be always on, the ideal would be dedicated servers + bandwidth on demand on a real-time bandwidth exchange. See http://www.invisiblehand.net

[Isabel Wang]

Hi Nemo,

Unfortunately the average hosting provider oversells its capacity to such an extent that bandwidth is completely devalued. It's not uncommon for a $10 hosting plan to include 2000 GB/month of data transfer. If a customer does the math, he'll conclude that each GB is worth half a cent - so if he buys in volume, maybe he can get the price down to a third of the cent. They'll be disappointed not to find such pricing on Invisible hand - unless hosting companies allowed customers to put their un-used capacity on the market. Which of course they won't. DreamHost calculated that if every customer used every bit of allocated resources, it'd cost them $200 to service a $7 account.

[Nemo Semret]

Hi Isabel, excatly. It's like mail-in rebates or gift certificates. They are designed with the assumption that a large percentage of people won't use them. It just means the true price is higher than the advertised price for those that don't, which is ok if they are accepting it as a price to pay for the extra convenience, but it can be misleading and lead to inefficiencies as is the case in bandwidth.