How Amazon S3 is going to change the world, 5GB at a time
How Amazon S3 is going to change the world was originally published in Web 2.0 Journal.
We are observing the transformation of the web from an ecosystem into an operating system. Building blocks such as websites, blogs, web services, podcasts and RSS are coming together and give rise to a new computing platform. The web operating system is emerging and it is bigger than the sum of its parts.
Remember your operating systems class, when you learned that every operating system has a handful of fundamental concepts such as storage, virtual memory and scheduling? The new web is no exception. However, since the Internet is a gigantic network of computers working in parallel, the basic operating system concepts take on a different shape.
For example, when you try to save a file on your computer, there is a (rare) possibility that the disk is full, and the write will fail. But with the new web operating system, this does not have to be the case. If the disk of one computer becomes full, the web operating system can switch and store the file on another computer. So on the web, you practically never run out of space.
Amazon S3 – the new virtual storage service from Amazon
The virtual storage has been in the news for sometime now. Dion Hinchcliffe has written a survey of virtual storage providers in his recent post. He particularly commended the Amazon S3 – simple storage service for its innovative API.
In essence, the Amazon S3 offers developers a huge hashtable. The minimalistic API, available in both SOAP and REST, is focused on basic management of the objects – write, read and delete. By default, the service works over HTTP and supports storage of objects up to 5 gigabytes in size. There is also support for BitTorrent and a plan to add other protocols in the future. To use the service, you have to have an Amazon Web Services account.
Amazon has done a very thorough job documenting and supporting the service. The resources page contains a wealth of useful information to get you going, most notably the API and the user forums. There are also code samples available in various languages that illustrate how to use the Amazon S3 API.
Storing and retrieving objects
The objects in S3 account are placed into buckets. Each account is allowed to have up to 100 buckets and the bucket name has to be unique across all S3 users.
Figure 1: Example from Amazon S3 API shows CREATE BUCKET request
Each object inside a bucket has to have a unique UTF-8 compliant key assigned by the developer. Since there is no specific key structure imposed by S3, the developers are free to do what best suits their needs. The documentation hints at using slashes to create directory-like structure, but does not insist on it. The lack of key specificity and directory interface in API is not a limitation, but an added flexibility, since people’s needs might be different and implementing the directory-like storage is just a matter of following naming conventions in the code.
Figure 2: Example from Amazon S3 API shows GET BUCKET request
The API also allows the developer to list all the keys in a particular bucket. This is implemented using a flavor of the Iterator pattern and a concept of a marker. With the first query no marker is supplied. If the bucket contains more objects than specified in max-keys parameter, then the a marker for the next starting point is returned. To obtain the next set of results, the marker is passed back in the subsequent request.
Figure 3: Amazon S3 diagram
Since it might be expensive to fetch the entire object from S3, the API allows you associate the meta data with each object. The meta is returned together with the key when GET BUCKET operation is requested. This functionality is particularly handy for storing and looking up information about large media files.
S3 has a built in security model for both connecting to the service and setting the access policy for individual objects and buckets. To access the service, each developer is required to obtain the Access Key ID and the Secret Key. The Access Key ID, the same ID used for all Amazon Web Services, has to be passed in with every request. The Secret Key is used as an encryption key to encrypt pieces of the request in order to prove the requestor’s identity.
The authentication is somewhat involved and there are quite a few questions on forums complaining about authentication problems. Amazon API has a page dedicated to it, which you can see here. In addition, there are code samples in various languages which illustrate the correct usage of authentication. If you do not know much about encryption, look for the code sample in your language – it will save you hours of debugging.
In addition to the authentication, S3 API comes with an Access Control List (ACL) for every bucket and every object. The ACL can be set via a separate request or at the time when the bucket or an object are created. Here is the set of currently supported ACL choices.
Figure 4: Current S3 ACL choices
Using AJAX to access S3
S3 opens an intriguing possibility of dramatically simplifying the back end for some applications. You can envision the architecture where an application simply consists of a client, which directly communicates with S3. This client can be a desktop application or an applet or an Ajax application embedded in the browser. Such a model is not appropriate for enterprise systems that require complex data processing, transactions and caching, but it could work well for things like Google Sync or del.icio.us. Simplicity, instant scalability, a built-in security model and Amazon’s reputation make a strong case.
It appears that the developer has plans to build this out further. It would be good to have a higher level of abstraction built in, as well as a way to loop through the objects in the bucket and ability to fetch multiple objects concurrently.
This all sounds very sweet, but what about the performance?
Needless to say, the performance is one of the top questions on the Amazon S3 forum. Amazon does not provide any specific data and does not have an SLA in place. But the design requirements and the design principles from the S3 creators show strong focus on performance and scalability.
There are a few benchmarks that independent developers posted on forums. I’ve seem these benchmarks improve since S3 launched. The latest results, indicate low latency. A benchmark on March 17 said: “Putting a 2 MB mp3 took 0.8s, retrieving it took 1.085s — quite fast, quite responsive”.
So who is using Amazon S3 right now?
Even though the service launched just a few months ago there is already a number of companies leveraging it for a wide range of purposes. In personal on-line storage space we note ElephantDrive (Windows), JungleDisk (Windows,Linux,Mac) and Filicio.us (Web-based). From the discussion forums it is clear that a few companies are planning to use S3 to store media files, but we cannot tell who these companies are. At adaptiveblue we are using S3 to store the bluemarks of our users favorite books, movies and music. And if you want to get a feel for what other things are possible, take a look at this top10 list.
Amazon S3 is an innovative, exciting new service that is going to change the way we do computing on the web. Michael Arrington of TechCrunch said this in one of his posts:
“S3 provides a terrific opportunity for startups with great ideas for a storage user interface to avoid building a back end storage infrastructure. Amazon is offering extremely low pricing and a very dependable infrastructure. For some people, S3 will allow them to launch a service that they otherwise couldn’t have built.”
If you have not done so already, take a look at S3 and see for yourself. A good place to start playing with it is the jSh3ll, which offers shell like command line interface to the service. Then join the Amazon Web Services program and start coding. The possibilities are endless!