Microsoft Windows Azure Cloud offers scalable resources to its customers. The price for renting the resources is linear, i.e. the customer pays exactly double price for double resources. However, not always all offered resources of virtual machine instances are most suitable for the customers. Some problems are memory demanding, others are compute intensive or even cache intensive. The same amount of resources offered by the cloud can be rented and utilized differently to speedup the computation. One way is to use techniques for parallelization on instances with more resources. Other way is to spread the job among several instances of virtual machine with less resources. In this paper we analyze which is the best way to scale the resources to speedup the calculations and obtain best performance for the same amount of money needed to rent those resources in the cloud.
Cloud Computing, HPC, Matrix Multiplication