Alleviating the Thrashing by Adding Medium-Term Scheduler - Advanced Operating Systems and Kernel Applications - page 129

Figure 4. a. Time Slice's Effect on Number of Swaps; b. Time Slice's Effect on Execution Time

as the results of Figures 2a and 2b. However,
the elimination of the thrashing saved more time
than was wasted idling, and the medium-term
scheduler still outperforms the traditional Linux
scheduler.
Figure 4a and Figure 4b show the effect of the
medium-term scheduler time slice on the process'
execution time. The tests were conducted using
SPEC. It can be seen that when the time slice
exceeds a certain limit, the execution time might
suffer. This damage is caused by the higher av-
erage idle time. When the number of processes
per group is too small, it may happen that none
of the processes in the current group is on the
Ready queue. Such a case may happen due to
many I/O operations. Clearly, this might turn out

with a lower group time slice as well, but it will
not happen as often as with a higher time slice,
because at the beginning of the time slice all the
processes are usually ready to run and not wait-
ing for an I/O.
When the time slice is higher, the cycle will be
longer. An extremely high time slice will actually
make the medium-term scheduler behave like
a FIFO scheduler. On the other hand, the page
faults rate is lower for the one-second scheduler,
because of the longer time slice. Pages are usually
swapped out when the group context is switched,
so if all the pages are replaced on context switch,
the half-a-second scheduler should have double
number of pages faults comparing to the one-
second scheduler. However, sometimes the bins