Research On Cognitive Decision Technology And Its Application In Constructing Reconfigurable Systems

Currently devices in the area of scientific computing can be classified into the general-purpose processor and the application-specific processor. The general-purpose processor can be applied to many areas but its efficiency is not satisfying while the application-specific processor is not flexible but of high efficiency. In recent years the reconfigurable computing device is becoming the buffer between the two kinds of traditional computing devices, which can not only be reconfigured into an application-specific device of high efficiency but also change from a function to another flexibly through dynamic reconfiguration. It is the high flexibility and reconfigurability that make the construction of reconfigurable system more complex to distribute its resources according to the needs of users than traditional computing systems. Therefore, a cognitive decision support system which can give proper decision at the right moment is needed. As a result, how to balance the resource allocation to realize the high efficiency and to satisfy QoS restriction of users needs the support of a cognitive decision support system. Based on this background, the main contents and contributions of this thesis are as follows:1. Based on the idea that the needs and loads of an application decides its efficient computing architecture, this thesis designs a cognitive decision model (CDM) based on Soar and realized the proactive decision support system (PDSS) which is applied in PRCA as an important component. This thesis summarizes the cognitive architecture techniques of cognitive decision technologies and analyzes the characteristics of the mature cognitive architectures both domestic and abroad, especially the cognitive architecture-Soar which is developed by the University of Michigan.2. This thesis proposes a SARM (Self-Aware Resources Management) managing method to cope with the problem of computing devices managing devised by our team of PRCA project. This method can manage the scientific computing devices effectively and intelligently and is fault-tolerant. This thesis presents an MTD-MPBR model (Multi Thread Distribute-Multi Process Block Reconfigure Model) to effectively reconfigure scientific computing devices. In addition, considering the reconfiguring effeciency and its impact on power consumption of the system, this thesis devises two greedy strategies allocating resourses-distribution and centralization. Then based on analysis of the two strategies, this thesis proposes a Multi-Object resource Allocation (MOA) algorithm considering both power consumption and reconfiguring efficiency.3. To decide the right percepting time in the cogntive decision model while construting the reconfigurable system, this thesis proposes a proactive perception algorithm ProPA (Proactive Perception Algorithm) in PULL mode which can guarantee the quality of the percepted information and effectively improve the effeciency of the reconfigurable system and its quality of service. In order to protect the system from exceptions between the two perception occasions, the information perception module adds a mechanism in PUSH mode to report exceptional information.4. In order to testify the proactive decision support system, this thesis devises and implements an English keywords extraction algorithm based on Impulse C. In the end, this thesis embeds this kind of application into PDSS and realizes an English keywords extraction system on the PRCA hardware platform. Experiments shows that the system performs better than traditional computing platforms in performance and power consumption.