AWS/400: Amazon Builds An AS/400-oid Cloud
April 23, 2012 Timothy Prickett Morgan
OK, there is no such thing as AWS/400, but conceptually speaking, the collection of 30 cloud services are the modern analog to the simplification and integration philosophies built into the AS/400 many years ago, all souped-up for a modern, Webby world. Or, at least that is what I kept thinking as I attended the AWS Summit in New York City last week. AWS is, of course, short for Amazon Web Services, and it is the cloud computing subsidiary of online retailing giant Amazon.
We have two main jobs here in the Four Hundred stack of newsletters. The first, of course, is to keep track of everything that is going on in the IBM i marketplace and provide coverage and analysis that is second to none. The other job is to keep an eye on the larger trends in the IT space and to see how these might affect IBM i shops, which do more than just run OS/400 and i platforms, and to try to encourage Big Blue, the steward and owner of the IBM i platform, to do this or that to better serve their OS/400 and i customer base. This particular story is one of the latter, with a bit of a thought experiment thrown in there for fun.
The first thing that struck me about the Amazon cloud is that it is now has roughly the same size installed base as the AS/400 at its peak, and once you adjust for a number of factors, roughly the same revenue stream. Let me explain.
Amazon is pretty vague about how many customers it has, but at the AWS Summit last week, Werner Vogels, chief technology officer at Amazon, who is in charge of the IT operations for the retailer and its AWS cloud alike, said that the utility computing and storage service now had hundreds of thousands of customers in over 190 countries worldwide. At its peak in 1998, the AS/400 division at IBM had around 275,000 unique customers and had about $3.3 billion in server revenues and $1.3 billion in internal and external disk array sales on the platform, for a total of around $4.6 billion.
As best I can figure, Amazon’s more than 30 different AWS services, which include its Elastic Compute Cloud (EC2) compute cloud, its Simple Storage Service (S3) object storage cloud, its Elastic Block Storage (EBS) block storage cloud, plus a bunch of database, networking, load balancing, firewall, email, and application framework services, are generating north of $1 billion in revenues. In 2011, Amazon had $48.1 billion in total sales, up 40.6 percent. Within this, a category called Other, which does not include merchandise or media product sales and which does include revenues from the AWS cloud, miscellaneous marketing and promotional agreements, other seller sites, and co-branded credit card agreements, grew by 66.4 percent, to $1.59 billion. I happen to think, and cannot prove, that most of this revenue is the AWS cloud.
Now do this thought experiment with me. Say Amazon is getting the impossible 100 percent utilization on its cloud here in 2012 to get something on the order of $1.2 billion in AWS cloud sales. In 1998, what was the average utilization on those AS/400 servers and storage that pushed that $4.6 billion in revenues? Was it even 25 percent utilization, averaged over the course of a week, a month, or a year? If the AS/400 was a cloud back then, then the incremental computing and storage capacity that IBM added in that one year was roughly the same revenue stream that Amazon had for its cloud in 2011.
To be fair, IBM had a much bigger installed base of capacity at that point, probably something on the order of 500,000 machines and, in this thought experiment, that might generate a cloud-like revenue stream of maybe five times this amount. Call it $6 billion for the whole installed base, if everyone could only buy the AS/400 capacity and storage they needed. If you adjust that for inflation, it is on the order of $8.4 billion in 2012 dollars. If the utilization was half what I expected, cut this in half. And honestly, I think that somewhere around $5 billion adjusted for inflation is closer for the aggregate cloudy value of all the AS/400 capacity in the world running in 1998–if it could have been fluffed up and cloudified.
My point is this: IBM started in the midrange with the System/3 in 1969 and took a decade after the launch of the AS/400 in 1988 to reach its peak in 1998. That’s 29 years. Amazon has built its AWS cloud in six years. And it is growing exponentially on the few metrics it provides.
For instance, the number of objects in the S3 storage cloud, which launched in early 2006, had 2.9 billion objects as 2006 came to a close, and passed through 102 billion objects three years later. As 2011 was coming to an end, the S3 cloud had 762 billion objects and as the first quarter of this year ended, there was 905 billion objects across all the Amazon cloud data centers stored in the S3 service, and it was fielding up something on the order of 650,000 object requests per second under peak load. By my math, the S3 cloud was averaging 1.6 billion new objects a day in Q1, and that was with automatic delete services running that Amazon added last year to keep S3 buckets from getting crufty. So just going linearly from here, you’re talking about S3 having about 1.34 billion objects as 2012 comes to a close, and obviously, on an exponential curve that S3 is on, it might be something closer to 1.6 to 1.7 billion objects.
It is hard for me to not be impressed by that. Amazon does not disclose how many users it has for the EC2 service or what number of virtual machines it has, but I am pretty sure those are on exponential curves, too.
But what I am really impressed by is that Amazon is understanding that building the cloud is not enough, that it needs to give away capacity for free (which it has been doing for several years now) to seed new customers and let developers learn about its platform, insulate developers and end user companies from the complexity of the AWS machine, and foster applications that can be easily and automatically deployed on its cloud.
Does any of this sound even remotely familiar to you?
IBM busted its tail with the System/38 and AS/400 designs to make the very tough thing of creating business applications that used a relational database as simple as possible, and eventually amassed an army of 8,000 application developers with over 20,000 applications to sell their wares on top of that platform, pushing IBM’s revenues. IBM used to make it very affordable for ISVs to get machines on which to write their code–which was as close to free as IBM could get with a physical machine that someone, somewhere has to pay for. (Amazon can just give you spare cycles it has so many, and boot you off when it needs them for paying customers.)
I could give you a lot of examples of how Amazon is insulating IT shops from IT so they can write code, but I will give you one example that will make this very clear to you. Amazon has been selling cloudy MySQL database slices for a while, called the Relational Database Service (RDS) and a non-relational data store called SimpleDB. The former is meant for transaction style systems, while the latter is really aimed at Web front ends and social media apps where immediate consistency is not a big deal. These are often called NoSQL databases to distinguish them from SQL-driven relational databases, which have belts, suspenders, bungie cords, duct tape, baling wire, chewing gum, and spit to absolutely ensure the consistency of the data pumped into a database table. These were good for what they are, but left something to be desired.
In January this year, Amazon launched DynamoDB, and it is a much better NoSQL database for a lot of technical reasons I won’t bore you with. Here’s the important thing: DynamoDB has two parameters that you specify–your read rate and your write rate–and that is it. You don’t worry about how much data you store, Amazon doesn’t care. It just sees what rate of database throughput you need and uses solid state disks on X86 servers to make sure you get exactly what you need. If you need more throughput for either reads or writes, you log in and dial it up and pay more. You can autoscale it, and put caps and floors on it too, just as you can with EC2, S3, and other services. Amazon takes care of replicating the data to multiple and redundant regions in its data centers so if there is a failure, you are not wiped out. High availability is based into the product from the beginning, just like relational processing was built into the System/38 and its RPG III compiler from the beginning.
NoSQL databases are crazy stupid complicated, and there are literally only two things Amazon makes you think about to use it. This is what makes AWS truly revolutionary. Not its virtualization and cloudiness, not its utility computing model, but the relentless desire by Amazon’s crack staff of techies to get companies out of the infrastructure management business. IBM is talking about automating the installation and patch management of systems, operating systems, application servers, and database software on the PureSystems machines announced a week ago, and Amazon has removed every bit of complexity from using a database. Your children or grandchildren will understand DynamoDB. Heck, I might even give it a whirl.
The point is, the Amazon cloud has gone from being a virtual computer slice to a virtual data center on which you can build real applications. It is a platform for the 2010s as much as the AS/400 was for the 1990s. For the moment, there are a lot of Web applications, as well as big data and high performance computing jobs and a slew of disaster recovery and archiving jobs. But real applications are making their way out there, too. One example that Vogels cited in his keynote was designed by Red Lion Hotels, which operates a chain of hotels in the American West with over 6,500 rooms in total–and its entire hotel reservation system is running atop AWS services. This is exactly the kind of customer that would have been running on an AS/400 two decades ago.
Also last week, Amazon jumped to the obvious conclusion that as an online retailer, maybe it should get around to it and create an online store and actually sell various software for the AWS cloud that can be paid for on an hourly basis, just like compute, storage, and networking, and be automatically installed to AWS slices. It’s called the AWS Marketplace and it bears more than a passing resemblance, conceptually, to the PureSystems Centre online store that IBM announced for its PureSystems. IBM has 150 apps in its PureSystems Centre online store; Amazon is starting with 250 apps. The race is on to build the catalogs and court the ISVs and end users.
I really do wish IBM would realize that IBM i is already a platform cloud of sorts, just waiting for it to be deployed on its SmartCloud so customers can get access to it and use it on a pay-per-use basis. If not, maybe we can get Amazon to roll in a few thousand Power System nodes and make use of DB2 Multisystem and other features to have Amazon build a truly scalable DB2 for i database service and build EC2-like images that work as RPG and COBOL compute engines that hit this database service. And if Amazon won’t do it, maybe Abacus Solutions, Connectria Hosting, or someone else will do it. This is more than just installing some PureSystem iron with IBM i installed.