Friday January 05, 2007

Predictions for 2007 You know, when I first joined Sun in 2003, there was a lot of talk about IT doesn't matter, and talk once again hyped up about IT commoditizing etc. So some say IT is important, and some say it is not, and of course the truth lies somewhere in the middle.

Every enterprise has it own unique business model, it;s way of doing business. Much of this business model is encoded in the form of programmes that can be executed automatically, or documented in procedures users will use when running those programs. This matters, this is what uniquely defines the business and maintaining this is essential to providing competitive advantage.

On the other hand, what I use to execute those programmes, doesn;t matter an iota. So IT doesn;t matter, if by IT you mean the computers and other devices that we use to execute our unique business model. But when we say it doesn;t matter, you can still get it very wrong. If you are running your unique business model on expensive proprietary components, your cost base will be far higher than if you are using commoditized components that you can purchase from a number of vendors who can compete for your business. Bryan Cantrill gives us an excellent overview of the economics.

The promise of utility computing is very simple - to do away with the wasted resources. Most data centres currently use 10 to 15% of their IT resources, the rest is wasted as machines sit idle, through the night and / or test machines that are seldom if ever turned on. The reason so much resource is wasted is that we can only do one thing on each computer, they take days to configure and software vendors will not guarantee that there application will run harmoniously side by side with anything else. The solution to these problems are now widely available, and we are starting to get to grips with what they mean to our business. The tools are Solaris Containers, which allows a single computer to be divided into an infinite number of virtual machines.

A typical architecture might include 3 web servers, 3 application servers and 2 database servers. Traditionally all of these would have been placed onto a seperate piece of hardware. Now using Solaris Containers we can place all 8 of these components onto a single computer, or maybe if we want more resiliency onto two computers. Now that solves part of the problem, we are now using the hardware a little more efficiently, but the rest of the problem is caused by the size of the computers you choose. Normally, we would have calculated the largest amounts of resources we might possibly require, and purchase hardware accordingly. Even with the containers above we will need to have a lot of power. What;s more we'll need a replica or two for performance testing, and another for development. So the key to getting real performance from our hardware is to be able to use it for multiple purposes, according to where our demand is at the time. This is not possible if we are still manually configuring servers, it needs to be a fully automated process, using tools like N1SPS and remote booting.

So this is all in place, what about the predictions. Well, it seems that the tools mentioned above, virtrualization rools, are going main stream and as enterprises learn to adopt them, they will be able to start levarging "Infrastructure as a Service". Infrastructure as a Service is something fundamentally different to hosting your servers at a provider, or oursourcing your managed operations, infrastructure as a service is about delivering the power you need, at the most competitive price possible. So how will the economics of it work? Well, enterprises are generally not going to rewrite all of their unique business model to use such a service, so we will see existing architectures being encapsulated in a virtualized world. That is, we will see people migrating their existing stuff on to a virtualised platform, such as solaris containers, and automating the deployment using N1SPS.

They key difference for Infrastructure as a Service, is that it will need to allow the customer to retain complete control over their business model, their code, and indeed over their data. I am starting to see customers who want to have their infrastructure deployed at two different sites, so they can fail over from one vendor to another. This is when you get best value, when there is nothing stopping you from moving to another vendor. It cannot be a project to move data centre, it needs to be at the flip of a switch.

So I'm expecting that we will see things like the Sun reference architecture programme moving towards producing standardised architectures based on containers, rather than specific hardware configuration, architectures that can be deployed on anything from a single small server, to 8 top end servers without change. I would expect to see these standardised architectures being offered inconjunction with managed operations, so that we might for instance have Oracle IaaS, SAP IaaS, J2EE IaaS etc. These will represent standardised offering that can be sold to support any number of users that can be provisioned onto any hardware. In effect Sun will be able to guarantee that the customers unique business model, be that stated in Java, .Net, SAP ABAP's, UML or whatever, Sun will be able to guarantee that it will run correctly, whereever the customer chooses to run it. Customers will then be free to shop around for who can provide the best hosting deal to support their needs.

In the longer term, I'm expecting those hosting options to be predominantly provided by the network operators, essentially standardised capacity will become a feature of the network they provide. In the meantime however, customers can look to providers like textdrive who are offering 1/16th of a server guaranteed, so our 8 servers above can be run on half a machine, and scale without downtime up from there. Larger enterprises might look to T-Systems who are also offering excellent dynamic infrastructure offerings.

The key here is that the customer can have a single consistent architecture, at a guaranteed price point, they can physically run their business model in different data centres, from different vendors, with no impact whatsoever on their day to day operations. Customers will retain control of their business model, using tools like N1SPS to deploy refined versions on a regular basis. The actual execution of the business model will be open to competitive tendor, according to who can acquire the power, space, cooling and network connectivity at the best possible price point. I would expect the billings to be made based on the reserved capacity for the entire system. Keys to look out for when discussing infrastructure solution with providers this year will be what are the units with which we reserve capacity - if they can only offer you box level you are in trouble. The ideal would probably be to reserve a number of clock ticks, possibly in Mhz or Ghz per service. Next thing to look to will be the service level to move the environment, if it takes more than 24 hours to increase the capacity of your system, you're in trouble. It should be possible to increase capacity of these systems while they are still running, that's why we architected them to have so many services, each service can individually be stopped and restarted on larger hardware without the system stopping at all.

Now you are in a world where finally you don;t buy as much hardware as you might possibly consume, but purchase as much power as you are actually consuming, this is the utility world, this is the type of commodity computing that doesn;t matter, it;s available to everyone and it;s about how you innovate on top of it. But if you take your eye off the ball in 2007, come 2008 and later, you'll find that your infrastructure doesn;t fit on to any infrastructure as a service and you are paying far more than you need to...

Anyways, none of this is official Sun position, but it;s what I personally expect to see happening over the coming year, it will be exciting to watch it unfold!


Posted by Gregory ( Jan 05 2007, 05:37:10 PM GMT ) Permalink

Thursday June 15, 2006

Hello World Hello World Yes - I was once a programmer, but it's been a while now. I used to describe myself as a "Clipper Guru" - if you remember that - but here in Sun that;s a little like bragging to Gary Kasparov that you are a snakes & ladders Guru! I'm a slightly strange fish in the world of computing, in that to be honest I don;t really like technology. I love what technology can do for me, when it works properly. I very much enjoy the automation of the mundane, but I don;t particularly enjoy tinkering with technology. Many years ago, at a trade fair where all of my colleagues had name tags featuring their various qualifications I added "QINAA" after my name. I was admonished by my management when they discovered this stood for "Qualified In Nothing At All". Still the only moniker I officially qualify for. For the last twenty years though, I have been implementing computer solutions from single user to tens of thousands of users. Almost all business applications you can think of. I have tried this from a number of perspectives from owning my own software company, to working for Global scale software vendors, hardware vendors, System Integration and consulting companies and end users. This variety of experience seems to count for something to Sun - they employ me as a "Solution Architect" and occasionally trust me to talk to very large clients on their behalf! Having said that - everything that I will be writing here comes straight from me - not the company line - not the company's opinion - just mine. Though I should warn you - I do buy the company line - I believe in the mission to cross the digital divide, I believe I am working for the good guys, I believe I can do my bit to make the world a better place by doing my job here at Sun. My role is part of our services practice. I'm part of a Global team supporting architects within our local sales organisations and interfacing back to our product marketing groups. This gives me the privileged position of meeting with great minds, inside of Sun and outside with some of the largest companies in the world. My particular area of specialty is "Utility Computing", whatever you deem that to be. I like to think of it as where the money meets the technology. This suits me well as it combines my years working the innards of accounting systems with my understanding just enough of the technology upon which they run! In the future I'll be planning to write about utility and whatever else crops up. Now let's see if I can figure out how to actually post this slightly long "Hello World" Posted by Gregory ( Jun 15 2006, 04:16:58 PM IST ) Permalink Comments [1]