There is a philosophical question: from engineering point of view, it is always better to have modular system with minimal connections between modules. I don't mention advantages of modular approach - let consider it as statement. The same time to achieve some additional gain and flexibility of the system, many developers (especially more mathematicians rather engineers) are looking for something called Cross-Layer Optimization, which is required more and more links between modules that moves from modular architecture to unified one: and all modularity is emulated as slices on software level. How do we can to merge these two contradictionary approaches into one, and do we need to do it ?
Interesting discussion... the main challenge lies in the complexity which will be imposed by cross-layer optimization. I think there is no fit-to-all answer to this question. If we want to proceed with a practical point of view the question should be more confined to a certain problem.
that is a problem that usually common solution is required for commercial market, otherwise it will be exclusive solution/architecture. Of course, solution can be fit some specific scenario by parameter adjustment. But it is big issue to adjust whole architecture to the specific conditions.
Another point: if system is too much complicated, then it is difficult to localize errors/bugs in the system without modularity.
PS it is clear that cross-layer is good for optimization, but still big question for implementation/debugging/robustness/functional replacement/fixed architecture, doesn't it ???
---> Himadri Nath Saha thanks for pdf, looks interesting...
Cross-Layer optimization does not mean hijacking or merging modules. Rather, it is creating another module which interacts with existing ones
If we need a low development price and few engineer headaches the answer is a modular approach (divide and conquer). Low investments and high regular costs.
if we want a low product price and high product quality we use cross-layer optimization (unit and conquer). In business terms we can say: "If we need very GOOD end-product we need big investment at the start, but we need large product series".
There is no contradiction. Just consider this as stages of evolution in time. At start a modular approach, when combining several modules (QAM+FEC or FEC+EQ) and finally the entire system.
P.S. Do not forget that the system consisting of optimal modules is not necessarily optimal by itself.
BTW, relevant work to the subject.
So, if we want to move into cross-layer design area, then need advanced algorithm development tools like CAE/CAD systems in circuit design. Now we can automatically design PCB with high efficiency, is there way to make similar design for algorithms - so called Computer-Aided Algorithm Engineering Processing ?...
I don't see modular design and cross-layer optimization as mutually exclusive concepts. Modular design has to be always there, the basis and starting point of the design. Once you build your modular design (software, HDL-based, HW-SW etc), it is easier to apply cross-layer optimization techniques. In fact modular design could be precisely considered as one of the enablers (or "assistants") of cross-layer optimization.
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity