Information Technology Reference
In-Depth Information
performance results. In the following subsections,
an extensive evaluation that has been made to the
medium term scheduler is described.
pf_ratio=1-((pfAMSQM-p)/(pfARC-p))
The TLB misses are shown as the ratio between
the TLB misses that AMSQM produces and the
TLB misses that ARC produces.When a page is
accessed at the first time, any algorithm will have
to induce a TLB miss and obviously there is no
way to eliminate this TLB miss, so we calculated
only the TLB misses of the pages just from the
second time they are accessed. The page faults are
shown also as the ratio between the page faults
that AMSQM produces and the page faults that
ARC produces counting for each page only the
second and further accesses.
tm_ratio and pf_ratio are the values that rep-
resent the calculation of the TLB miss ratio and
the page fault ratio respectively.
testbed and benchmarking
We implemented the standard CLOCK algorithm,
the ARC algorithm and the AMSQM algorithm.
We used Valgrind (Nethercote and Seward, 2007)
to capture the pages that were used by some of the
SPEC - cpu2000 (SPEC, 2000). The SPEC manual
explicitly notes that attempting to run the suite
with less than 256Mbytes of memory will cause
a measuring of the paging system speed instead
of the CPU speed. This suits us well, because our
aim is precisely to measure the paging system
speed; hence, we simulated a machine with just
128MB of RAM, although it is obviously a very
small memory.
The sizes of the Super-pages that we used were
8 KB, 16 KB, 32 KB, 64 KB, 128 KB and 256
KB. We assumed a tagged TLB of 32 entries for
instructions and 64 entries for data.
Both AMSQM and ARC outperform CLOCK
by all the parameters in our simulation, so we
found no point in presenting the results of CLOCK;
therefore, the results presented here are only the
ratio between strict ARC and AMSQM.
Let us define:
benchmarking using Spec-2000
Figure 1 and Figure 2 show the extra overhead of
ARC over AMSQM. Figure 1 shows the tm_ra-
tio of several selected SPEC2000 benchmarks
whereas Figure 2 shows the pf_ratio of the same
SPEC2000 benchmarks. It can be clearly seen in
Figure 1 that AMSQM achieves a higher TLB
ratio, because of the super-pages usage.
Furthermore, AMSQM memory hit ratio is
also higher than ARC memory hit ratio in most
of the benchmarks as can be noticed in Figure
2. The improvement of the memory hit ratio is
because AMSQM takes advantage of the locality
principle as is mentioned above in the introduction
section. The other SPEC benchmarks show similar
results, so we do not include these benchmarks
in this paper.
n - Number of memory requests by the
benchmark.
p - Number of pages that the benchmark
accesses.
tmARC - Number of TLB misses when ARC is
the replacement algorithm.
tmAMSQM - Number of TLB misses when
AMSQM is the replacement algorithm.
pfARC - Number of the benchmark's page faults
when ARC is the replacement algorithm.
pfAMSQM - Number of the benchmark's page
faults when AMSQM is the replacement
algorithm.
tm_ratio=1-((tmAMSQM-p)/(tmARC-p))
the threshold Setting
considerations
Figure 3 and Figure 4 show the influence of
threshold on the system performance. Too high
threshold harms the TLB hit ratio, whereas too
low threshold harms the page fault ratio; hence, it
Search WWH ::
Custom Search