Information Technology Reference
In-Depth Information
dynamically determines the state of the database items and adapts the broadcast
contents by monitoring the broadcast misses. As a result, the “hot” data items are
being pushed and the pull-based data items are being unicast based on individual
requests. Similar to the states of water, the database items are classified into three
groups: Vapor, Liquid, and Frigid. Vapor represents the class of the data items that
are hot and therefore, they are being broadcast. However, due to the passive nature
of the clients, the server does not receive any feedback regarding these data items
and hence, gradually they will cool down and turn into liquid. Liquid represents
the class of data items that are directly requested by the clients and as a result they
are being unicast. These data are either being requested more and more and after
passing a threshold level become hot, or they are not going to be requested and
hence freeze. Data in frigid state either stay in that state or migrate to the liquid
state.
3.1.2.2 Adaptive Push-Pull.
Bhide et al.
[8]
proposed a scheme that,
based on parameters such as the network condition, scalability, degree of coherency,
and resiliency to failure, intelligently and adaptively determines the degree of push
and pull data that are disseminated.
A Time-to-Refresh (TTR) value is assigned and calculated by the proxy for each
cached data item. The TTR determines the next time that the proxy should poll the
server in order to refresh the cache contents. The TTR is computed based on the
rate of changes of data and the degree of coherency required by the client. A smaller
TTR shows a rapidly changing data and/or strict coherency requirement. A larger
TTR, on the other hand, represent infrequent changes and/or relaxed coherency re-
quirement. Two algorithms have been proposed, namely Push-and-Pull (PaP) and
Push-or-pull (PoP).
In PaP, the proxy is responsible to pull changes to the data, and the server is
allowed to push additional updates undetected by the proxy. In PoP, the server is
allowed to adaptively choose between push and pull-based dissemination schemes.
3.1.2.3 Dynamic Leveling.
Dynamic leveling scheme
[55]
is a natural ex-
tension of the hierarchical broadcast model discussed in Section
3.1.1.3
. Recall that
the hierarchical model allocates frequently accessed data items on faster broadcast
channels. However, the access frequency is dynamic and time dependent. The pro-
posed dynamic leveling algorithm adjusts data items among the broadcast channels
when the access frequency changes. Experimental results have shown greater effec-
tiveness of the dynamic leveling algorithm relative to the VF
K
algorithm, in term
of execution time when the database size and the number of broadcast channels in-
crease.
