| With the wide spread of computers'applications and the improvement of technology, people realized that the processing ability of a single processor is quite limited. For the single processor, as one main part of a single computer, it also makes the processing ability limited and makes it far away from satisfying computing requirements. As a result, people naturally think of a way of using several computers to construct a multi-computer system. But how do the computers in the constructed multi-computer system cooperate with one another? It is a problem concerned with dividing a job into tasks that can run separately and assigning the tasks to computers. Central queue algorithm, task-assignment-by-guessing-size algorithm (TAGS) and extended TAGS algorithm are of the algorithms proposed to deal with the problem of assigning tasks mentioned above. This article describes in details the central queue algorithm, TAGS algorithm and extended TAGS algorithm. The latter one is based on the former two algorithms, absorbs the advantage of the two and have better performance. The main contribution of this article is that it elaborates the implementation of extended TAGS algorithm in Redhat Linux operating system with kernel version of 2.4.18, and designed a task flow, in which the length of the tasks follows the heavy-tailed distribution, to input into the resulted system to test and validate its design, and finally measured the mean slowdown of the system. As for the TAGS and extended TAGS algorithm, the theoretical calculation of their performance is based on the assumption that the arrival process of the task sent to one server is Poisson arrival process. The actual process is not Poisson but hard to describe in mathematics. The work done by this article gives convenience to the using in practical system of the central queue, TAGS and extended TAGS algorithm, the performance comparison of those algorithms based on experiments and further study of those algorithms and other algorithms concerned. |
PDF Full Text Request