So far, we have only seen the performance benefits of compressing the swap area. As we mentioned in the introduction, this was the main objective of the project. Anyway, we also had a second objective that consisted on increasing the number of pages that could be placed in the swap area. A study of this objective is presented in this subsection.
Although the number of pages that fit in the swap partition depends on the compression ratio, it is not the only important factor. The fragmentation found inside the buffers will also be an important parameter in the final size of the swapping space. A system that leaves large unused portions in the buffers will not be able to place many more pages than the original system in the swap partition.
Figure 7 presents the size of the swap area that we would obtain if we could fill it with pages following the same compression ratio and the same fragmentation as the ones obtained in the benchmarks. In the figure, we can see that in most cases the system increases the size of the swap partition more than a 50%.
Figure 7:
Size of the swap area that would be needed to have the same capacity
as our 128Mbytes compressed swap.
We have limited the gain to 256 Mbytes as we have configured the size of the virtual_swap_info table to double the physical swap space. If a greater array were used, a larger swap space would have obtained with benchmarks such as simulator and xanim.