Autonomic computing: How we manage the complexity
Q: One of the issues that comes up with having a ubiquitous virtual computer running as part of the Internet is that you are also creating an inherently complex system. Another area of investigation is autonomic systems, which it is purported will create a more self-healing environment for computing. Can you give us a status report on autonomic computing?
A: Let me first comment on the complexity question because this is very important. Like other major technology infrastructures, such as electricity or the telephone network, the aim is that even though the infrastructure itself is complex--say to generate electricity you have Hoover dam, transmission lines from Hudson Bay, nuclear power plants in Canada--if want to toast a bagel in the morning, you don't have to know any of that.
It is important to encapsulate the complexity so that users of IT and developers of applications don't have to deal with it. They simply have a set of interfaces that they use and those will work very nicely.
Now comes the question of how do we manage this complexity. One of the major ways that we absolutely have to develop to manage this complexity is to automate it. We have to make the systems themselves much more self managing; to build into our systems and the whole infrastructure the ability to detect failures and heal themselves; to detect attacks and protect themselves from hackers or other attacking systems; to optimise themselves so if have an application that is slowing down because of too much demand, let it on the fly add computing capacity so it continues to provide a reasonable response time level; to configure themselves so when you add a new piece equipment to the system it automatically detects it, brings it online and everything works.
To do that we launched an initiative in IBM last year called autonomic computing. We launched it as an industry-wide initiative because we felt these are such complicated problems that all of us in the industry need to cooperate to make everything more self-managing.
What progress has been made so far in getting cooperation from other companies?
A number of companies have developed their own autonomic computing projects, such as HP and Unisys. I think that, as often happens in the industry, when someone sees a good idea, the markets follow. We are all exchanging research ideas. For example our Almaden research labs in San Jose California, ran an autonomic computing meeting in early April. We had participants from Stanford, Berkeley University of California, Sun, HP and other companies in the industry all exchanging ideas at the research level.
An area where we will collaborate in building autonomic infrastructure will be around grid computing precisely because in order to build a heterogeneous, self-managing infrastructure, you need to have a set of common protocols that run on every system so that the various systems can collaborate with each other.
You will see us, as a community, collaborate more on security that works across various systems and on self-healing algorithms. For example, workload managers that run across the infrastructure that can detect which nodes are having problems and therefore should be taken out of line and that work routed to other nodes that are operating well. Because of heterogeneous infrastructure we need common protocols. Building autonomic capabilities is one of major application areas on top of grid protocols.
'Grand challenge' problems demand more sophisticated systems
Many companies, like IBM, are developing point solutions such as workload managers. But the goal was to create something more like the human nervous system in terms of automatic response systems in which things happen quite naturally. What is the gap between building these kinds of point solutions solving discrete problems like workload management and fail-overs to something more sophisticated, more artificial intelligence and control theory?
Especially when you want those autonomic capabilities to work across the infrastructure, those are very difficult problems. Those are right now in the research stage. Again, we have to agree on a common set of protocols, which is part of what has driven us together around this grid community. We have to work with universities and research labs around the world to develop it.
These are very challenging, sophisticated problems. We will be able to do simpler things first. There is quite a bit you can do in workload management, such as detecting with heartbeat controls on all the nodes of the infrastructure. Other things like being able to do a very good job detecting denial-of-service attacks and take actions will take longer because there has to be sophisticated pattern recognition work to tell difference between a surge in usage and a true denial-of-service attack. You will see progress at different levels. I would call these Grand Challenge problems, meaning these are definitely complex problems that will take quite a while to get right.
Nanotechnology, which has the ability to create machines atom by atom, has been getting a more press lately. When do you think nanotechnology will be practical? Within the next 10 years?
A lot of this depends on the definition of what we mean by nanocomputers. We are seeing computers get smaller and more sophisticated. We are entering the next stage of computers called blades, which are nowhere near nanotechnology, but they are significantly smaller than even today's computers and will let us put together much larger numbers of computers in clusters.
We will see that every year we learn how to build smaller and smaller components and how to integrate them closer and closer together until eventually--probably 8-10 years--we will be approaching a level that feels more like nanotechnology, but a lot of it will depend on the definition.
We have projects at IBM research. For example the Blue Gene project is aiming to put together a million different computers interconnected in a special network to achieve petaflops of computing power in particular to tackle very sophisticated life science problems that are ahead of us, such as genomic and proteomics research. If you look at what is going on in storage technologies and the incredible densities that one is able to achieve, we are almost reaching the levels where quantum effects begin to matter. We are beginning to reach those points and people have to get much more sophisticated in building computers, but I view that as an evolutionary step.
