Adaptive computing, self-healing systems, Grid and on-demand computing, autonomic computing.... Vendors from all sides are throwing buzzwords around, a new one every day or so it seems.

This month we'll try to make sense of it all by looking at what is here today, what will be here tomorrow, and what is mere science fiction. More important, we'll examine how these new ideas impact your ability to develop and deploy applications.

In tough economic times, IT managers are asked to do more with less. That means driving down the cost of everything from laptops to enterprise-wide applications. The cost of deploying an application lies primarily in two areas: hardware and people. Standards-based software platforms such as BEA WebLogic have been instrumental in reducing cost and time-to-market of applications by providing a reliable base and a set of built-in services.

Increased Complexity
Standard platforms also allow a much higher degree of application interoperability by allowing applications to talk to each other using standard protocols like RMI, JCA, or SOAP (Web services). That move from silo architectures to interconnected, high-level architectures has increased the complexity of applications considerably. From client/server we've gone to Web client-Web server-business tier-database, with the business tier potentially connecting to many other applications, including legacy systems and Web services across the Internet. We have Web site objects, workflow objects, database connections, and JCA connections. The platform has kept up with this increased complexity, but only barely. To take applications to the next level a new set of technologies that radically simplify the development, deployment, and maintenance of applications will be required. We group these technologies under the label "Adaptive Computing" because in that model the infrastructure adapts itself to the application. It optimizes, provisions, and heals itself without the intervention of developer or administrator.

From Grid to Adaptive Computing
Grid computing is among the first ideas for the sharing and optimizing of computing resources, and comes straight from the halls of academia. The idea behind Grid is to take some large chunk of work, say the mapping of the human genome, and break it up into many small chunks spread over many computers. The Search for Extraterrestrial Intelligence at Home (SETI@home) project at Berkeley is an example of this. Its goal is to analyze radio telescope data to detect specific patterns indicative of extraterrestrial intelligence. The amount of data collected is so enormous that no single computer could possibly do it by itself, so researchers devised a scheme where anyone with an Internet connection and a PC could participate by running a screen-saver program capable of analyzing a small chunk of data. When your computer is idle, the program uses the idle CPU cycles for the project. To date, the SETI@Home project has had 4,257,524 users who contributed a total of 1,336,810.852 years of computer time. So far no one has found an extraterrestrial. (Note: This may or may not be entirely true. At least one signal matching the target profile was recorded by the Big Ear radio telescope at Ohio State University on the night of August 15, 1977. It was never detected again.) Since then many other projects have emulated SETI to solve hard scientific problems, from breaking cryptographic keys to finding a smallpox vaccine.

More recently, the Globus project has developed the Globus Toolkit, an open-source implementation of a Grid infrastructure, written in C. The toolkit is a "bag of services" that can be used to develop Grid applications and programming tools. While some companies are talking about using Globus in an enterprise setting, Globus is really designed for the scientific and engineering problems we just described rather than the problems found in corporate IT.

Think Globally, Act Locally
Like many academic ideas, Grid needs to be refined before it can be used in the real world. The vast majority of businesses, enterprises, and government organizations don't want to spread their data or their applications all over the Internet, or even across computers they don't control completely. While interacting with computers and services on a different network or across the Internet is common practice, sending one's applications and data is not. IT professionals want to maintain administrative control over their IT infrastructure. IT departments want to make the most efficient use of their hardware and don't want idle CPUs. The solution is to evolve and broaden the Grid to the more powerful concept of Adaptive Computing. Adaptive Computing is an umbrella term for a far more intelligent application infrastructure. Such an infrastructure makes better use of resources through dynamic provisioning, self-healing, and self-tuning.

Better Provisioning
IT departments must often allocate enough machines to handle peak demand for a particular application, leaving most of their boxes idle most of the time. Traffic at e-commerce sites such as Amazon.com or FedEx may be highest in the weeks leading up to Christmas, but lowest after New Year's. A CRM application may peak during the day when customers call in while the inventory application could make use of the same hardware at night, when no one is calling in. Upcoming versions of application infrastructure will let applications share hardware and other resources effectively to minimize duplication and hardware costs.

While saving on hardware costs can generate large savings, development and maintenance costs dominate the cost of deploying an enterprise application. Companies like Microsoft and BEA have focused on reducing the cost of development with tools like BEA WebLogic Workshop and Visual Studio .NET. The cost of testing, optimization, management, and administration, however, is still too high. This is where so-called "self-tuning" and "self-healing" applications can save an enterprise a lot of money.

Imagine if you will a system that notices that its process performance is slowly degrading over time. After running a diagnostic procedure, it concludes that one of the applications running in the JVM is leaking memory. At that point it will notify an operator and take action by itself: it may quiesce the application in the question process (i.e., instruct the application not to take any new request, complete all outstanding requests, and shut down) and leave the other applications alone and the process running. It may quiesce all applications and either 1) restart the process minus the offending application or 2) restart the process with a bigger heap, until the application is fixed. The BEA WebLogic Platform provides robust self-healing features, including fail-over and automatic connection pool resizing, but this is just the beginning and you'll be seeing much more coming in that area.

While self-healing is the ability to deal with exceptional conditions gracefully, self-tuning is about improving the application's performance under normal conditions. In other words, self-tuning is the ability for the platform to optimize itself for a particular application. Platforms such as BEA WebLogic have hundreds if not thousands of configurations and tuning knobs. Today a typical application is tuned in a testing lab by a developer with a load simulator in one hand and a tuning guide in the other. A developer or an administrator can adjust many parameters, including memory heap size, the number of execution threads, the number of IO threads, the size of EJB caches, or the size of a JMS queue. The idea behind self-tuning is to let the infrastructure monitor the application, gather and analyze the data, and based on that data optimize the application automatically. This has the twin benefits of making the infrastructure easier to use and of improving application performance. As for self-healing, the BEA WebLogic Platform has been leading the pack with self-tuning features. Its J2EE JDBC drivers, the software that lets Java applications connect to databases, have long been self-tuning. There again there is much more we can do.

Easier Deployment
The idea behind dynamic provisioning, or the sharing of hardware resources, is to treat a large pool of computers (a distributed system) as if it were just one computer, much like a mainframe. We call this the "Virtualized Mainframe." BEA WebLogic has pioneered the most advanced and robust implementation of the key concepts needed to do this, such as clustering, load-balancing, and fail-over. There is much left to do however, and you should look for some exciting improvements in the coming years. These include distributed application deployment so that deployment, undeployment, and quiescing of applications across a domain becomes seamless. Application containment is included so that a specific application can be granted a specific amount of resources, but not more. This is key to ensuring that no one application can take down an entire IT domain, either by mistake through a programming error, or by design through a virus or a Trojan horse.

Reducing Complexity
All of these features have one aspect in common: automation. Automation, or letting the infrastructure do more and the administrator do less, is the only viable way to reduce complexity. Managing, optimizing, understanding, and debugging the applications of the future will only be possible through radical simplification. This is what Adaptive Computing is about.