Information Technology Reference
In-Depth Information
CPIX considers two important factors: local
access frequency and the global availability of
data objects. Access frequency helps to identify
critical objects that should be locally cached to
improve local cache hits (and hence reduce wait-
ing time and energy consumption). Global data
availability is used to identify the data objects
that are difficult to get from outside, i.e. they are
neither widely cached by other mobile peers nor
frequently broadcast by the server.
Although CPIX considers local access interest
rather than neighborhood's access interest, in fact
it benefits the whole peer community when every
mobile peer is doing this. When every mobile peer
is caching the data objects that are hard to get
from external sources (including neighbors and
broadcast channel), the global data availability
is improved.
Both the information of access frequency and
the information of global data availability are es-
timated from local statistics. A mobile peer may
estimate its access frequency to the data objects
by counting historical accesses. The global data
availability of its interested data objects can be
estimated from the waiting time it experienced
for the local cache misses. To do so, each mobile
peer keeps the following statistics for every data
object D i it has accessed:
The CPIX scheme works as follows. Each time
when a query for data object D i arise:
1.
The mobile peer increases NumAccess i by
one.
2.
If the query results in a local cache miss ,
the mobile peer increases NumLCM i by one,
records the time point when the local cache
miss happens, and broadcasts a request m e s -
sage to neighbors. If KR (Keep Requesting,
see the Background section) is enabled, in
this step the mobile peer will keep sending
out request message every certain time in-
terval if neighbor cache miss f o r D i happens,
until the mobile peer gets D i .
3.
When the mobile peer gets D i from either a
neighbor or the broadcast channel,
a.
The mobile peer records the time point
and calculates the waiting time for this
local cache miss of D i , and add the
waiting time to SumWaitingTime i .
b.
The mobile peer decides whether to
cache the data locally. If there is free
cache space available, the mobile peer
caches D i using the free space. If its
cache space is full, a score for D i is
calculated. If D i 's score is not the
smallest among the scores of cached
data objects, the mobile peer caches
D i by replacing the data object with
the smallest score. The score of a data
object D i is calculated using the formula
as follows.
• NumAccess i : number of access.
• NumLCM i : number of local cache miss .
• SumWaitingTimei: i : the sum of waiting time
the mobile peer experienced due to the local
cache misses of D i .
SumWaitingTime
Score
=
NumAccess
*
i
i
i
NumLCM
Notice that the idea of global data availability
(viewing the data availability at other mobile peers
and the data availability on the broadcast channel
as a whole) simplifies the caching scheme. In CPIX
a mobile peer does not need to get the knowledge
of data objects' global availability by exchanging
summary of cache content with neighbors and by
analyzing the broadcast index.
i
The rationale behind the score formula is
that Score i tells the expectation of waiting time
caused by a local cache miss of data object D i .
NumAccess
NumAccess
We use
i
as the estimation of the
access probability to data object D i . Since every
 
Search WWH ::




Custom Search