How not to curb Java Enterprise Applications

How not to curb Java Enterprise Applications

Modern enterprise servers running Java applications typically feature between 16 and 64 cores, and can additionally be bundled in clusters or grids. Old fashioned applications written for sequential hardware will simply not be able to make use of this hardware. What is required is a Java Parallel support for running tasks concurrently on a server or across servers.

When different tasks are totally independent, to the point of not having to communicate with each other, making Java parallel amounts to have a potentially large number of JVMs running independent copies of the same application. When tasks need to communicate, there are two major options: sharing memory, or exchanging messages.

Sharing memory typically requires tasks to run on the same server, in different threads of a same JVM. This is the approach that requires the less changes to existing code, but is difficult to scale because all tasks share the same memory bandwidth. This is the case handled by the standard Java parallel library. Other types of parallel deployment will require different tools.

Some tools try to simulate shared memory across a network of computers. This approach sounds appealing at first glance, since it suggests that traditional sequential Java programs may be deployed with little change on a network of computers, or even in the cloud. The drawback is that simulating shared memory relies on a protocol of message exchanges that are invisible to the developer but generate heavy traffic. In short, simulating shared memory over a network is slow.

The other approach is to use explicit message passing for communicating between concurrent tasks. Transforming a Java application to use message passing can in some cases require an important architectural change, but it also provides much better insight into the app's behavior (what specialist's call "parallel thinking") and can be finely tuned for the best performance. The most important feature of a message-passing Java parallel application is that it scales easily over a large variety of hardware configurations.

Scientists and engineers have been doing intensive calculations on supercomputers for decades, developing applications such a physical simulation and weather forecasting. These high-performance applications have been designed in message-passing style, using dedicated tools from the C or FORTRAN world. Now it's time for the much larger ecosystem of Java Parallel applications to benefit from similar advances.

POST COMMENT

Network Marketing Secrets, Guide to Wealth in Seven Steps Annihilation Deck by Cameron Francis Revealed (free download) Complete HD experience at Home with DISH Network Fiverr Marketplace - Why Fiverr Is An Internet Marketer's Dream and Can Save You Money Dynamic Web Site Content is a Must in Today's Internet Marketplace How Do I Secure My Network Tips To Help You Fix This Computer Wireless Internet Freezes Internet Marketing - How Much Should You Charge for Your Professional Service? What Is A Network Marketing Internet Business? How Can You Get Hold Of Best Internet Deals? Internet Security, Some Useful Tips DISH NETWORK Enhances Business Prospects of its Patrons in USA Enjoy Top American Channels on DISH Network