Swap Token - Advanced Operating Systems and Kernel Applications - page 94

Figure 3. The memory performance of gzip and vortex in their concurrent execution.

A process gains more memory pages and in-
creases its RSS through page faults. On the other
hand, it loses pages when these pages become old.
In this way the global page replacement policy
attempts to make the memory allocated among
multiple processes to conform their respective
memory demands.
Unfortunately, what a process loses includes
false LRU pages, which are generated during its
period of faulting. The losing of these false LRU
pages does not reflect the memory demands. This

study shows that the proportion of false LRU pages
in all the page faults keeps increasing with the
increase of memory shortage. Consequently, the
dynamic memory allocations are hard to reflect
the memory demands of processes. For example,
gzip established its working set during the period
of time between 600 th second and 760 th second,
because we observed that its page fault rate is
significantly reduced. Then some of its memory
allocation was transferred to vortex , illustrated
by the lowered gzip RSS curve and increased

Figure 4. The memory performance of bit-r and gcc in their concurrent execution.