A General Nash Equilibrium Semantic Cache Algorithm in a Sensor Grid Database - Data Management in Cloud, Grid and P2P Systems - page 21

1, 33% of Restaurant: 33% of Hotel: 33% of Packing lots;
2, 50% of Restaurant: 30% of Hotel: 20% of Packing lots;
3, 70% of Restaurant: 20% of Hotel: 10% of Packing lots;
4, 80% of Restaurant: 10% of Hotel: 10% of Packing lots.
We consider two main performance aspects:
(1) Average Response Time. The response time is defined as the elapsed simu-
lation time from the moment that a query is issued at a mobile device M org to
the moment that M org obtains all of the answers.
(2) Average Peer Hit Ratio. A storage hit occurs when a desired semantic cache
block is found, if the system hasn't retrieved it from the server. We use the peer
hit ratio for the percentage found in peer storage.
5.2 Experiments
Because the strategy of using CSC is obviously poorer than the other ap-
proaches, we mostly compared the following three different strategies: SCSTL ,
B & B GDSF , CSDS .
We chose two parameters to compare the simulation result, which are Seman-
tic degree: The Semantic degree takes on the values of 1 , 2 , 3 ,and 4.
(1) Average Response Time. In Fig. 3, when the Semantic degree is 1 the ap-
proach CSDS becomes similar to our approach SCSTL . For all other situations,
the Average Response Time of our approach SCSTL was always obviously bet-
ter than the others.
(2) Average Peer Hit Ratio. In Fig. 4, in the beginning, when the Semantic degree
is 1, the approach SCSTL is poorer than the others. However, when the Seman-
tic degree increase to 2 and 3, the eciency increases. When the Semantic degree
reaches a max value of 4, our approach SCSTL attains the best eciency of the
Average Peer Hit Ratio. In daily living, people will choose some hot places, which
means that our approach can attain the best eciency in a realistic application.
Moreover, generally speaking, in the beginning, the Average Peer Hit Ratio of
our approach SCSTL is not as good as the others, but as time goes on, its
eciency increases, and in the end it obtains the best result, which means that
our approach SCSTL is specifically suitable for a long-term application.
Furthermore, higher Average Peer Hit Ratio, means less communication and
computation cost. Especially in realistic applications, communication distance
between peer to peer is much less than between peer to server, maybe 100 times
or more, so it can save lots of bandwidth and energy. Moreover, it is obviously
that computation and query costs in peer are much less than in server, maybe
1000 times or more, for the cache data quantity is limited. So that, in realistic
applications, SCSTL can attain much better eciency than others.
The reason for all of the better e ciencies of the Average Response Time and
Average Peer Hit Ratio is that our approach SCSTL synthetically considers all
of the semantic, time and location factors.