Dissertation > Excellent graduate degree dissertation topics show

Multi-level program cycle behavior analysis

Author: YangXiaoXi
Tutor: ZhangWeiHua
School: Fudan University
Course: Computer Software and Theory
Keywords: Cycle Behavior Analysis Multi-level Forecast Exactness
CLC: TP311.1
Type: Master's thesis
Year: 2011
Downloads: 6
Quote: 0
Read: Download Dissertation

Abstract


Periodic behavior analysis is done by comparing the characteristics of the different procedures block information analysis program similar approach. Assigned to the same period as the behavior of the fragments have similar performance characteristics and resource needs, therefore, periodic behavior of dynamic optimization has been widely used in various systems, such as dynamic cache reconfiguration, the dynamic cache prefetch optimization based on feedback information, Dynamic energy optimization, acceleration and reduce software simulator debugging instrumentation overhead and so on. In the periodic behavior analysis, cyclical behavior of granularity effect cycle predictive accuracy as an important factor, but not before there specifically for its research. Most of the previous cycle behavior analysis methods are based on fine-grained, these methods documented and constantly updated during program execution cycle behavior information and use this information to predict future periodic behavior. This only consider the most recent local periodic behavior while ignoring the overall periodic behavior, resulting predicted periodic behavior due to the local variation of localized noise, thereby reducing the accuracy of prediction. This is the first carried out on the multi-level program cycle behavior analysis possibilities were studied, this approach combines the fine-grained and coarse-grained behavioral analysis cycle periodic behavior analysis to improve prediction accuracy. A coarse-grained block is usually caused by dozens or even hundreds of block composed of fine-grained. Our study found that: ● belong to the same periodic behavior of different coarse-grained comparison between blocks, they contain fine-grained periodic behavior of the composition and distribution of high similarity and stability; ● When a coarse Granularity block contains a number of fine-grained block former belongs periodic behavior is known, can be more accurately determine the current block belongs grained periodic behavior. Based on these findings, we designed and implemented a multi-level program cycle behavior analysis (MLPA) method. In the MLPA method, coarse-grained block cycle to determine the behavior of the program is based on a paragraph to begin a number of fine-grained block. The coarse-grained block the rest of the fine-grained block before the adoption of periodic behavior of the same kind of coarse-grained record periodic behavior in the behavior of fine-grained cycle sequence to make predictions. This method is purely cyclical behavior based on fine-grained analysis of the major advantages compared to consider comprehensive periodic behavior of the program, thereby overcoming consider only the most recent information on the behavior of local cycle brings one-sidedness. Experimental results show that this method can obviously improve the cyclical behavior prediction accuracy. Relative to the Markov prediction method, MLPA forecast for the next cycle, the cycle change prediction and forecasting cycle length were increased by 17%, 42% and 32% prediction accuracy, but only a 2% increase time overhead and 40% (approximately 360 words section) of space overhead. In order to verify the effectiveness of this method, we first applied to the MLPA dynamic cache reconfiguration systems. The system can dynamically adjust the size of the data cache to reduce energy consumption and access times. Experimental data show that, compared and fine-grained, MLPA used during program execution can average 15% reduction in size of the cache. Secondly, we will MLPA applied to a sample-based simulator accelerated systems. The results of the accuracy of the simulation results in the case of considerable, MLPA strategy can shorten the simulation more practical. Compared with 10M SimPoint using MLPA strategy can achieve 14.0 times (15 hours vs.125 hour) acceleration.

Related Dissertations

  1. Research on the Evaluation of Straw Resource and the Development of Scale-Up Straw Energy Utilization in Jiangsu Province,S38
  2. Study on Growth Monitoring Technique Based on Pixel Un-Mixing Method and HJ Remote Sensing Images in Paddy Rice,S511
  3. Research on Talents Team in Hefei-Wuhu-Bengbu Based on Independent Innovation,C964.2
  4. TV weather programs \,G222
  5. Different treatment methods agar gel replication complex mode of phosphate suitability of,R783
  6. Zhaoguan Lower Coal Group water inrush prediction and control techniques,TD745
  7. Research on the Feasibility of Highway Construction Project,F542
  8. Analysis of China’s Household Electrical Appliance Industry Supply Chain Integration,F426.6
  9. Research on Methods of Fingerprint Image Segmentation,TP391.41
  10. Safety Assessment of Coal Gangue Hazard and Analysis of the Greening Reclamation,TD849.5
  11. Applied Research of Ground Penetrating Radar in Tunnel Geological Prediction,U452.11
  12. Research of Theory and Application on the Traffic Demand Prediction on Highway Feasibility Study,U491.14
  13. Modeling of Photovoltaic Cells and Its Application in Energy Prediction,TM914.4
  14. A JIT Inventory Model for Supply Chain Management,F253.4;F224
  15. Gray comprehensive evaluation based real estate development risk assessment,F224
  16. Multilevel Finite Element Analysis for Femur,R318.01
  17. Characteristics Analysis for DNA Sequences of the Influenza Virus Based on the Time Series Theory Methods,R346
  18. Analysis and Forecast of Health Care Expenditure and Its Constitution Based on Panel Data Model,R197.1
  19. Climatic Features and Statistical Prediction about Sea Fogs in 3 Islands of Yellow Sea and Bohai Sea,P732
  20. Based on multi -level stations Jinchuan two mine ventilation Ventilation systems analysis,TD724
  21. The Region Prevention Technology of the Coal Spontaneous Combustion in the Large Goaf,TD752.2

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer software > Program design,software engineering > Programming
© 2012 www.DissertationTopic.Net  Mobile