Matrix multiplication performance analysis in virtualized shared memory multiprocessor

In this paper we analyze the performance of a multiprocessor using a virtualized environment. There are a lot of papers analyzing the performance with conclusion that discrepancy happens due to L2 cache. Our conclusion based on observation and experimental research shows that the main reason should be L3 in this case. Our hypothesis to be confirmed by experimental research is that the program will be executed slower in virtual environment compared to traditional server. There is also a hypothesis that the speedup in virtual environment for parallel execution is smaller than in sequential. However, analyzing the speedup graphs we can conclude that speedup in virtual environment is greater than speedup in traditional environment. We have also found that there is a region with particular problem size for matrix multiplication where the program runs faster in virtual environment. The experiments showed that the achieved performance in virtual environment is even better than traditional environment, but only in specially defined regions. In other regions there is a huge performance drawback in virtual environment and fails to measured average of 66.58% of traditional environment.