OS as the Operations Research (OR) Internet
There is one fundamental difference between the current NEOS and Optimization Services. NEOS is based on a tightly coupled centralized structure. All the solvers are connected with the server, and all the optimization job requests have to go through it. Therefore, the system does not scale well.
On the other hand, Optimization Services adopts a decentralized Service-oriented Architecture. There is still in some sense a “central” server in the middle, but it functions as a lightweight “registry server,” or just “registry.” Such a registry knows all the solvers and other Operations Research software that exist in the whole decentralized system by keeping metadata files. Metadata here means that the registry contains information about the software, but not the software itself. No solvers are actually executed by this registry; instead users directly contact the solvers in a peer-to-peer mode. The advantages of a decentralized Service-oriented Architecture are significant and are elaborated throughout this thesis. The Internet has become popular because it is a decentralized architecture. There is no such thing as a “central repository server” that hosts all the Web pages. Development and maintenance all happen spontaneously. It is our vision that a decentralized architecture can better promote research and development in Operations Research. In Section 2, we discuss the Optimization Services process, which illustrate the close analogy between the Optimization Services and the Internet architectures.
|