Information Technology Reference
In-Depth Information
▪
1
=|B
2
|/|B
1
|
P=Min(P+δ
δ
the importance of the last access time to a larger
page is less weighty.
◦
1
,C)
Move the page from B
◦
1
to be the LRU
of T
2
(swap out page according to P).
If the requested page is in B
fragmentation
•
2
:
◦
If |B
2
|≥|B
1
|
δ
ARC and LRU do not take into their consider-
ations the location of the “victim” that is chosen
to be swapped out; therefore, they can leave many
holes within a super-page, because some of the
base pages that the super-page consists of can be
in the memory whereas some others can be out of
memory. When not all of the base pages are in the
memory, a promotion can be costly. A better page
replacement algorithm must check the “neighbor-
hood” of the victim page. A similar scheme was
suggest by Romer et al., 1995), but the authors of
this paper have preferred to use LRU.
▪
2
=1
◦
Else
▪
2
=|B
1
|/|B
2
|
P=Max(P-δ
δ
◦
2
,0)
Move the page from B
◦
2
to be the LRU
of T
2
(swap out page according to P).
•
If
the
requested
page
is
not
in
T
1
∪T
2
∪B
1
∪B
2
:
o Move the new page to be the MRU of T
1
(swap out page according to P).
As we mentioned above, CLOCK can move
its clock hand over many pages, until a page with
an unset bit is found. Unlike CLOCK, ARC has
a constant complexity - O(1). In addition, ARC
is tunable i.e. ARC can adapt itself according to
the characteristics of the data that the processes
use. These are the reasons why we chose to adapt
ARC to the Super-Paging mechanism.
thrashing
Usually supper-pages supported systems are less
thrashing-proof than the old traditional paging
systems. They cause a more extensive memory
consuming that can lead to a memory pressure
and even a thrashing (Wiseman, 2009), (Jiang,
2009). Obviously, when the memory is very large
this deficiency is not critical. However, one of the
importance advantages of the ARC over the “re-
cency” algorithms is the thrashing-proof feature;
hence the thrashing argument is clearly in favor
of the use of ARC.
Super-pageS and arc
When adapting ARC for super-paging environ-
ment, some considerations should be taken into
account. In this section, we would like to discuss
these considerations and to see how they can affect
the ARC algorithm.
coarse granularity
The use of super-pages causes all the super-page's
base pages to be considered as important (or
unimportant) pages, whereas usually just several
base pages are important or sometimes even just
a single base page is important. In such cases, the
LRU algorithm gives poor results, because it has
no mechanism to distinguish between super-pages
containing many important base pages and super-
pages containing just a small number of important
larger pages
When using super-pages the pages that are used
by the operating system are usually larger. In
such a case the recency is less important than
the frequency. If a page is frequently accessed,
it can hint the operating system that this page is
important, even if the page is very large, whereas
Search WWH ::
Custom Search