Laboratory for Internet and Innovative Technologies

CPU cache is used to speedup the execution of memory intensive algorithms. Usage of greater cache memory sizes reduces the cache misses and overall execution time. This paper addresses architectures in modern processors realized as multi chip and multi core processors with shared L3 cache and dedicated L2 and L1 cache. The goal is to analyze the behavior of servers and to check the hypothesis if the virtual environments are usually slower than standard environments. Although majority will think that adding a new software and operating system layers just slows down the software, there is also a contra-argument based on cache size exploitation, since a multiprocessor with dedicated cache per core usually requires smaller cache size and can exploit the benefits of bigger cache memory than single processor. This might also lead to superlinear speedup greater than the number of processing elements. The testing methodology and experiments for this research are applied also to cloud environment. The results show that cloud environment can also achieve superlinear speedup for execution of cache intensive algorithms when high performance computing is used in virtual machines allocated with more than one processing element (core).


Sasko Ristov, Magdalena Kostoska, Marjan Gusev, and Kiril Kiroski


. . .

Full Paper

The paper is published in BCI 2012, Proceedings of 5th Balkan Conf. in Informatics, Novi Sad, Serbia, ACM International Conference Proceedings, ISBN 978-1-4503-1240-0, 2012