While deploying distributed components, a key decision to be made is the location of each component in the target distributed environment. Unappropriate distribution may lead to bad performance. Existing distribution algorithms use criterion such as minimal communication bandwidth or response time to optimize object distribution. But for distributed applications processing massive parallel requests, which require both low response time and high throughput. In such situation single-criterion algorithms may output distribution with a very low throughput sometimes. We propose an algorithm called OCDA based MINLP(Mixed Integer Non-Linear Programming), which meets the requirement that with a restricted average response time, maximize throughput capacities of applications. Finally we discuss the advantages and limitations of OCDA.
展开▼
