Amazon Web Services are readying their latest service called EC2 which will allow users to setup and run servers and computing capacity in the cloud. Users of the service can setup a server instance which is hosted with Amazon, and then access and use the servers they setup just like any other. With EC2 there would no longer be a requirement to source and setup physical hardware and the virtual server instances are charged back to the user based on the CPU, storage and bandwidth usage.
The pricing of EC2 is 10 cents per instance hour (which comes to $72 per month for a server that is always available), 20 cents per GB of bandwidth and 15 cents per GB of storage (storage is with S3). Compared to traditional server providers such as ev1servers this may not be priced low enough (especially the bandwidth cost, considering most hosting providers include 2000GB or more of bandwidth) but it may prove to be a good solution for some users.
The way it works is that you use tools that Amazon provides to create a machine image on your local machine (the tools are all written in Java). You can setup the image with a web server, application environments, mail or anything else, the “images” are just Fedora Core and they come with some pre-installed services (Amazon calls them AMI’s, or Amazon Machine Instances).
To setup your own server instance you then upload this image to Amazon S3, and then once it is uploaded you go to Amazon EC2 and register the image as a server instance. Once registered, you can boot and access the server instance within minutes. Each server instance is the equivalent of a 1.7Ghz Xeon CPU, 1.75GB of RAM, 160GB of local disk, and 250Mb/s of network bandwidth. Currently EC2 will allow you to create up to 20 server instances, to create more you need to contact Amazon.
While each server instance provides decent computing power applications such as large-to-medium scale databases or large web applications will require work to bring the computing power together to serve requests. Since each instance has a fixed amount of capacity they will be prone to performance issues when under heavy load as achieving scalability requires the user to acquire more server instances. One issue is that having separate server instances is not true “elastic” computing, like what Sun or other vendors provide, since the user is responsible for clustering and or load-balancing solutions between the servers.
For users who need smaller scale solutions then EC2 would work well. I would say that it will only be a matter of time before we see some front-end providers pop up and offer “instant on” servers pre-configured for tasks such as web and mail hosting. How billing would be handled by these providers is an unknown, but this could open up the lower-end shared and virtual hosting markets since it can provide the end user everything they need where they pay only what they use for. The biggest advantage this service has is that it is a very quick way to setup a new server online, and you pay for what you use, starting at just the base rate for server time (10c per hour).
No word yet on when this service will be opened up to the general public, it is currently being trialed by a small number of long-time Amazon Web Services customers. Full documentation and forums are available on the site now.
Um, are you sure that’s the correct Fedora link?
The service is pretty easy to set up – I was able to get a server up and running in a few minutes. Screenshots at: http://www.cast...zon-ec2-review/
Re: Nik’s comment on per server performance, I imagine it wouldn’t be long before Amazon or other developers distribute server images with built-in clustering software. Then the advantages of EC2 would be obvious as startups create cost-effective mini google-scale clusters: $7200/month for a cluster of 100 machines.
Power to the people. The net will continue to revolutionize the way we do business. Web 2.0 is just getting started. It is bigger, better and stronger the web 1.0. I hope that everyone has something to offer this revolution.
/*
Anyone who has lost track of time when using a computer knows the propensity to dream, the urge to make dreams come true and the tendency to miss lunch.
–Tim Berners-Lee—
*/
Holy @#$@… We were about to move to Rackspace mainly for uptime support. I’d imagine this would just never go down as well (in spite of Amazon.com’s recent downtime, S3 stayedup). We provision servers at 10GB of storage and 500GB of bandwidth, so we’d only pay about 180$/server, give or take.
That’s what we pay now at a medium-tier provider.
Definitely sent an email to the tech team. Thanks for the heads up on this Nik!
Hmm my verdict is still out on this service.
Great post, Nik, although you gotta be careful when claiming exclusives: Chris Pirillo posted at 3am, and I’m sure there were others before him.
http://chris.pi...pute-cloud-ec2/
This kind of scalable computing resource is especially interesting to those creating small web services which may or may not take off. You have a bunch of things converging here — many new, small sites are getting done with frameworks like Ruby on Rails. These fameworks attack the problem of lowering initial development cost, while EC2 lowers initial hosting cost. Result? Much lower barriers to experiementation and market entry.
Pirillo had it…
I love this. Between this and S3, Amazon is moving up on my dreamy companies list.
I would gladly use this for new projects as opposed to renting a dedicated server. Even if you couldn’t get as much power out of the individual machines, the combined power of all the servers you could get with your dollar is pretty attractive.
Yow! Fedora link updated, don’t know what I was thinking there. Thanks Mark
Pete: I started writing this post long ago when it was an exclusive, I hope Mike doesn’t notice I was late
updated the headline anyway
As for server images with built-in clustering, that might just work.. but there is a lot more to running and maintaining an application that spans more than just a single server. I think a truly elastic computing solution would allow the user to just keep on adding more power and memory to it
I wonder what it’s based on. vmware xen or what?
hmmm my dream has come true
The Amazon EC2 Limited Beta is Currently Full
BOOOOHOOOOO, I hate techrunch for this
Techniques for scaling web applications over multiple server instances are relatively well understood and accessible to developers of moderate skill.
What is less available is the ability to scale to multiple server instances at costs like amazon is offering. I would exepect that the availability of hosting at this pricepoint will drive the creation of frameworks and techniques to make such development and deployment even easier.
This is not to say that it wouldn’t be “better” to be able to scale a single server instance elastically, but I don’t think that is a prerequisite for innovation in this space.
I do think amazon needs to provide some facility for load balancing & failover for incoming requests though, before this thing can really take off, either that, or they need to allow people to provision the network topologies required to do it themselves (if they did, a 3rd party could offer a load balancing front end service to other users, built on amazons infrastructure).
Actually, this makes me wonder, what is the cost of communicating between two E2 server instances “owned” by different entities. Is it free?
According to Amazon’s info, data transfer between EC2’s or EC2 and S3 is free! See http://www.amaz...?node=201590011
Joris: It is based on Xen. See the 5th screenshot on http://www.cast...zon-ec2-review/ . Both xenwatch and xenbus are running.
I don’t really understand what this is. Is it hosting? Can I put a php + Mysql app up there?
pwb –
Yes you could use this as a webhost, however it’s not really worth it if your not using a webfarm. However if you ARE, you can make your site have the smarts to add webservers when your load gets too high.
pwb – As far as anyone on the outside can tell, yes it is. Internally, there’s a lot more to it, but the wonderful part of this whole thing is that it’s completely transparent to the user. I never expected this from Amazon, of all companies.
Also, I lolled at “Almost Exclusive”.
As for server images with built-in clustering, that might just work.. but there is a lot more to running and maintaining an application that spans more than just a single server. I think a truly elastic computing solution would allow the user to just keep on adding more power and memory to it
Note that Amazon doesn’t consider these servers “reliable” in the sense that they state in their introduction to the service that data stored directly on them or used in memory should be considered transitory. They suggest storing data permanently by moving them off the computer into other structures. I’ll be curious if you could set up MySQL, for instance, to host its databases at S3 — doesn’t seem like that would quite work because you need a socket. Amazon database services–on-demand SQL–would be fabulous.
What impresses me the most is the holistic approach that Amazon is taking. They are carefully crafting these building blocks individually, but also, thinking about how they fit to gether to deliver collective value. This is going to be an amazing, as in A-Z (pun intended) platform for delivering the software of the future.
Alex
Amazon is competitive with teir 1 providers. A full rack and a hundered megabit link is around 7500 a month if you go teir 1 all the way around. Amazon is around 9K if you were to duplicate the same configuration. You could easily recoup the 1500 bucks in saved operating costs and trips to the data center.
EC2 remedies one cost inconvenience of S3. Let’s say you put 20 terabytes of unindexed bulk data into S3. Then, say you need to scan it for a string (or very similarly, index it). Just scanning over 20TB would cost ($0.20/GB * 1024GB/TB * 20) $4096 in bandwidth at $0.20/GB, be limited to public-internet speeds, and you’d still need to provide the scanning/indexing machines.
With free bandwidth from S3 to EC2, the bandwidth and machine costs disappear. If a machine could scan at just half the 250Mb/s wire speed, it’d take under 16 days to scan 20TB and cost under $38 (!) — less than 1/100th the previous bandwidth-only cost.
But more temporary instances seem costless to provision, so use 400 instances for 1/400th the time, and get the same 20TB scanning job done at the same cost — in under 1 hour.
Wow.
“A full rack and a hundered megabit link is around 7500 a month if you go teir 1 all the way around.”
Don’t know where you’re getting pricing, but our typical pricing is about 60% of that price, and it’s Tier-1, premium 100% uptime bandwidth all the way around. I know our pricing is fairly competitive with other colo companies out there, and if you want to go cheapo bandwidth, it’s less than half the price you are quoting.
Amazon’s price isn’t competitive, but I don’t think it’s intended to be. This is for tiny companies that are testing things out. Most established companies are still going to want to be able to see their servers and have full access to them 24×7 through the use of KVM-over-IP, etc.