Enhancing the Efficiency of Memory Management in a Super-Paging Environment by AMSQM - Advanced Operating Systems and Kernel Applications - page 289

apsi, crafty, bzip2 and gzip, executed in
parallel.

ARC. The significant improvement in the TLB
ratio of AMSQM comparing to ARC can be simi-
larly explained in the other traces.
Figures 8a and 8b show the page faults of
AMSQM vs. to ARC. It can be clearly seen that
AMSQM achieves a higher memory hit ratio in
all the benchmarks, because of a good utilization
of superpages and based on the locality principle.
However, the improvements vary from 0.2% for
trace 4 up to 10.4% for trace 1. We found out that
ARC performs efficiently in trace 4 (and trace 3)
i.e. does not produce many page faults, because
there is enough space in the main memory, and
therefore AMSQM's improvement is relatively
small.
Yet, we found it very encouraging that for an
extreme heavy memory consumer benchmarks

Figures 7a and 7b show the TLB miss ratio
of AMSQM vs. ARC. It can be easily seen that
AMSQM TLB misses are significantly fewer than
ARC TLB misses. It can be noticed that trace 2
achieves a higher TLB hit ratio comparing to strict
gzip. This can be explained as a result of the TLB
coverage in this experiment which is significantly
smaller than the TLB coverage in the previous
experiment; thus a base page replacing algorithm
such asARC will experience enormous number of
TLB misses, whereas an algorithm suchAMSQM
that utilizes wisely the Super-paging mechanism
will gain a higher TLB coverage and hence will
produce relatively less TLB misses comparing to

Figure 8. a. First group of heavy traces page faults; 8b. Second group of heavy traces page faults