Amazon EC2 Beta

Innovation never takes a break, and neither do I. From the steaming hot beaches of Cabo San Lucas I would like to tell you about the Amazon Elastic Compute Cloud, or Amazon EC2, now open for limited beta testing, with more beta slots to open soon.

Amazon EC2 gives you access to a virtual computing environment. Your applications run on a "virtual CPU", the equivalent of a 1.7 GHz Xeon processor, 1.75 GB of RAM, 160 GB of local disk and 250 Mb/second of network bandwidth. You pay just 10 cents per clock hour (billed to your Amazon Web Services account), and you can get as many virtual CPUs as you need. You can learn more on the EC2 Detail Page. We built Amazon EC2 using a virtual machine monitor by the name of Xen.

Amazon EC2 works in terms of AMIs, or Amazon Machine Images. Each AMI is a pre-configured boot disk  --  just a packaged-up operating system stored as an Amazon S3 object. There are web service calls to create images, and to assign them to virtual CPUs to run your application. If your application consists of the usual web server, business logic, and database tiers, you can built distinct AMIs for each tier, and then spawn one or more instances of each type based on the load.

In a previous post, Sometimes You Need Just a Little..., I alluded to the new world of scalable, on-demand web services. In that post I talked about the fact that sometimes a little bit of storage is all you need.

Sometimes you need a lot of processing power, and sometimes you need just a little. Sometimes you need a lot, but you only need it for a limited amount of time. Perhaps you are doing some number crunching, some in-depth text processing, some scientific research, or your end-of-month accounting. Or perhaps you want to experiment with some radical new parallel processing algorithm for a week or two. In any of these situations, acquiring sufficient hardware to accomodate the high-water mark of your usage would definitely not be economical. There are already some interesting examples of this in the Amazon EC2 Discussion Forums. For example:

• Daniel Drucker says "We're planning on using it for functional MRI analysis. We have large datasets which, when they're being processed, require a cluster of 15-20 machines... but we only need those machines for a couple hours every few weeks."
• In the same thread, a user by the name of spanglu says "Let's say your back-office inventory app is web-based, but that you're only using it from 7am to 7pm.  You can cut your server costs by 50%. Take this to its logical conclusion - only start up an instance when you actually need your inventory app..."

Put another way, time is another interesting axis of scalability.

Before the advent of Amazon EC2, you had to buy or rent sufficient servers to cover your present needs, and you also had to be able to anticipate, forecast, and pay (in advance) for enough hardware, storage, and network bandwidth to accomodate organic growth as well as bursts of traffic brought upon by popular sites such as Digg or Slashdot. If you are too generous with your planning, hardware sits idle. Too frugal, and your chance at fame and fortune may very well pass, as thousands of would-be users are greeted with a "site too busy" message.

With Amazon EC2, you don't need to acquire hardware in advance of your needs. Instead, you simply turn up the dial, spawning more virtual CPUs, as your processing needs grow. During the beta you can run up to 20 virtual servers per account, or more by special arrangement.

Returning to our hypothetical three-tier application, Amazon EC2 gives you the ability to control network access on a very fine-grained basis. For example, you can allow the outside world to talk to the web servers, but not to the business logic or to the database server. You can allow the web server to talk to the business logic, and the business logic to talk to the database, and that's it. You also get free, fast-path access to Amazon S3, making S3 a natural place to store your raw data and your results.

Ok, developers, now that you have access to this computing resource, how can you take advantage of it?

For starters I would recommend that you get your hands on a copy of Cal Henderson's new book, Building Scalable Web Sites. Cal was one of the engineers behind FlickR, and his real-world experience in building a high-traffic site are aptly recounted and generalized in his book.

Second, start looking at higher level packages that will let you decompose immense computational tasks into a form suitable for parallel processing. It is no secret that one of Google's tricks is a software framework that they call MapReduce. This framework simplifies the task of performing similar processing steps on millions or even billons of pieces of data. Doug Cutting's open source version of MapReduce is probably worth investigating, as is Starfish, an implementation of MapReduce in Ruby.

Third, consider what you can do to help other developers use Amazon EC2. What about building specialized AMIs and then selling them to other developers? Preload an AMI with a popular open source stack (being careful to respect any and all redistribution prohibitions in the software licenses). What about an advanced monitoring system that spools up additional machine images in times of heavy load, then safely winds them down after the load goes away? There are a lot of ways that you can add value on top of what's already there.

Finally, the phrase "leveling the playing field" is used quite often with regard to the full line of Amazon's Web Services, but I think that it is truly appropriate here. I find it worthwhile to imagine that there is a developer (maybe its you) sitting in a dorm room or a spare bedroom somewhere, burning the midnight oil and creating the next killer web site or search algorithm. Without services like Amazon EC2, Amazon SQS, and Amazon S3, you'd need to invest a lot of capital, max out your credit cards, re-mortgage your house, and take a pretty big financial risk just to see if your ideas will work in the real world. With these new services your risk is a lot smaller, but your potential reward is in no way diminished.

That's about all I have to say. You might find Jon Udell's S3 screencast interesting, and you can always find something entertaining on the requisite Slashdot thread.

-- Jeff;

Comments

[Glenn Fleishman]

Thanks for the tips on where to read how to distribute tasks. One of my weekly tasks is that a dedicated $2,500 computer (two years old at that price, so probably $1,000 of current computational power) loads a large dataset and spends about 24 hours using close to full processing power to parse and normal data into a few gigabytes worth of linked MySQL tables.

I need to reconceptualize the task, as I imagine splitting it among even 2 computers would be difficult at the moment, since all the tasks are dependent. If I could turn a 1-computer, 24 hour task into a 20-compute 3 hour tasks, that would cost me a grand total of $6, and probably a couple bucks in terms of moving the databases on and off my production systems, and storing a backup in S3.

[Leitsystemseiten]

A quite intrestingidea is realized in this website! And a good and easy to handle design has been found too!

[Fred Domke]

As exciting as this is from a cost/benefit perspective, I think the real value to my customers in B2B integration is the elimination of the remaining barrier to building effective and efficient networks that extend business processes across enterprise boundaries has been eliminated. The Virtual VAN already leveraged Amazon’s S3 for virtual storage and SQS for the Internet point-of-presence. Now we can build an image of EME and deploy it to EC2 and eliminate the need to install and configure a server at all! All an enterprise will need is a lightweight connector that sends and receives messages to SQS. Everything else is ‘virtual.’ As the barriers are removed, the full business value of B2B integration will finally be realized.

[Danny de Wit]

It's seems like a great service that enables small firms, like ours, to launch their new products on a great infrastructure. Which translates into a very big relief of pressure for us.

If the service is available when we launch, our two new on demand products in development (a personal productivity tool + calendar & a enterprise backend system), will probably both be on EC2.

Any idea when this gets out of beta? Or when it will allow more users? We've signed up too late, but would love to launch on it.

We expect to go live in 2 - 3 months time.

[Interesting]

Nobody stores all their money in a mattress, they trust a bank to store it, and just have their identity ready to claim it. Similarly, I think the days where everybody saves their data, or hires IT administrators and buys servers and all the attendant headaches are going to be history in the next 10 years. Small businesses, freelancers, consultants will particularly benefit from this paradigm change, and cut out all the middlemen and overhead. Online services and renting compute power on demand are absolutely the wave of the future.

If someone at Amazon is farsighted enough to assemble a "virtual company template" which integrates a ticketing system, elastic compute cloud, a CRM and customer database, version control, backup, standard company website template with an integrated product download and payment system, and a tax/expenses template, there will be a exodus of cubicle dwellers itching to create new products. Amazon could well emerge as the new IBM-like services giant, and Microsoft will be left far behind. But the above requires execution and vision, and a lot luck in being at the right place and the right time. Time will tell.

[Sheina]

The term "Cloud Computing" emerged after 2007. However, Why did Amazon already call their Service "Elastic Computer Cloud" back in 2006?