the chest of memories came to receive the documentation of my final project. I was surprised to reread the introduction to my work (a computer simulation framework applied to cache coherence in NUMA machines) and I clearly see what they say: I am of letters and a mole in the technology world. I'll post the introduction of three fragments. To put this into context, let us consider that I wrote back in 2005, I have not changed a comma and then it was probably a lot smarter now ...
The first of these solutions would be to overcome the disabilities of computing power and speed of the fastest processors possible to build (and cheaper) combining the skills of many simple processors. Thus, computing clusters has allowed the treatment volume surprising information and respond to unmet needs of power before. Would for example, computing clusters rendered images, modeling weather or stock. And as crude version of such clusters would appear the load-balancing clusters that share the work between various machines semiparalela. For example in the case of web server farms that share the responsibility to meet the requests of the customers' browsers.
other hand, the redundancy of some of the sensors and processors of human organisms inspire other types of clusters. Clusters of high availability or fault tolerant, able to cope their duties with tolerable loss of performance even after a failure in one of its elements.
But the clusters have managed to generate a new computational model comparable to that of the Turing machine? The answer is no. Finally, improvements have been made (not always) at the speed of execution of complex algorithms, but in theory there is nothing to prevent a Turing machine to perform the same task. Does this mean that this line of research is dead and useless? Here each of the parties to the dispute give their own version of events.
as technology clusters would undeniably useful and practical. On the one hand, freeing the computing capacity of the machines of progress in building processors with more capacity and higher cost. The power to solve a task with items cheaper and more scalable systems has obvious advantages in the real world application of information technology. So in this sense does not seem likely cease to exist, this type of machine or leave it for research in this line.
However, from the theoretical point of view we have seen that this type of equipment has revolutionized the science as some expected and that finally brings no new computing model. What has been the failure? The initial idea based on observation of human beings and their complex parallel processing did not seem wrong, the question is what are the computer clusters or parallel machines really only give an illusion of parallelism? If the algorithms implemented on a cluster computer can theoretically (without limitation of time and memory) to run on a Turing machine, it is clear that parallelism does not exist and is merely an illusion of parallelism equal to that already existed in the machines uniprocessor running multithreaded applications and multi-parallel giving an impression of unreality. However, the research should not be abandoned as invalid, because there is no evidence that is incorrect, the results achieved are not yet satisfactory.
This search for a new parallel computing model really has not been abandoned at all. The new concepts provided by relativity in the physical sciences and experimental studies have come to open up new scenarios for this new parallel model. Quantum computers can address a new computer model in which computing states would no longer be 0 or 1 binary and exclusive but could coexist simultaneously. In this sense (moving parallel and could not be otherwise so) the newest theories in psychological science begin to model the mental processes of the same way: with simultaneous states. That is, with real parallelism.
So, in my view the evolution of computing as a science will present identification and realization of this new computational model. On the other hand, computing and technology will continue to address, given its advantages, and for quite some time building and improving multiprocessor machines or clusters.
0 comments:
Post a Comment