Research On Scalability Of Multi-core SMP Cluster Parallel Computing

With the rapid development of high-performance computing technology,parallel computing has become the main method to improve the performance of computer systems.And scalability has become one of the main metric to evaluate parallel computing performance.With the scale of high-performance computer system increasing,however,there exit many issues,such as more complicated structure of parallel computing system,difficult programming,large requirement of storage capacity and low efficiency.The above problems reduce the performance of parallel computing from different aspects,thus they have a certain impact on the normal operation of the parallel system.It can be seen that the performance of parallel computing does not increase with the increment of the machine size.Nowadays scalability has become an important research issue in parallel processing,which is used widely to describe whether the parallel algorithms(parallel programs)can utilize scalable processors effectively.Therefore,depth and comprehensive study of the scalability of parallel algorithms and architectures is the key for the reasonable and accurate evaluation of parallel computing performance.In view of the above,the thesis mainly studies the scalability of multi-core SMP cluster system.Scalability metric is the basis of the scalability of parallel computing.We should establish appropriate scalability metric to realize parallel computing performance improvement goals.The main contents of the thesis are as follows.First of all,the thesis introduces the architecture and model of parallel computer,as well as the architecture of multi-core SMP cluster system.Then we introduce the concept of scalability and related technology in detail.Based on the classic scalability metrics,the thesis introduces the range and characteristics of their application.Besides,according to the characteristics of multi-core SMP cluster system architecture,the traditional scalability metric of ISO-efficiency can be improved.The processor set we analyze is not exactly equivalent in the multi-core SMP cluster system.In order to solve the problem,this thesis defines the computing power of the processor set based on the concept of optimal processor set.It improves the traditional scalability metric of ISO-efficiency to fit the multi-core SMP cluster system.And the improved scalability metric is analyzed.It is concluded that traditional metric is a form of improvement metric under certain conditions.And we obtain the conclusion about the relationship between improvement metric and parallel execution time.In addition,on the basis of improving the traditional scalability metric of ISO-efficiency,we further study how to adjust the inter-node scalability of the cluster system.In order to makefull use of the incremental processor as much as possible in the parallel algorithm,we choose the appropriate network interconnection structure.Increasing the bandwidth of network communication can reduce the communication delay and calculation delay effectively,thereby improving the performance of the system.Therefore,the concept of scalability adjustment factor among nodes is proposed based on bandwidth.And the scalability metric is further optimized.In theoretical and experimental way,we analyze that the scalability adjustment factor among nodes on the system plays an important role to enhance the performance.Finally,the thesis introduces the dawn TC5000 as the parallel computing experiment.The experiments about extension are carried out on cluster platform by running the program for the matrix multiplication algorithm in order to further validate the effectiveness of the improved scalability metric.We make the detailed analysis of the experimental data.The experimental results show that the improved scalability metric is a good way to analyze the scalability of multi-core SMP cluster system.It is significant to improve parallel computing architecture and to tune parallel algorithm design.