QoS-Oriented Grid-Enabled Data Warehouses - Data Warehousing Design and Advanced Engineering Applications - page 161

Figure 4. General view of the SLO-aware scheduling model

a certain deadline ( mlet ) that is specified by the
Community Scheduler. When a local scheduler
agrees to execute a query by a certain deadline,
it makes a SLA (Service Level Agreement) with
the Community Schedule.
When a SLA is signed, the local scheduler is
committed to execute the query by the negotiated
deadline. But it can, for instance, reorder local
query execution (do not have to execute each
of the incoming queries as fast as possible) or
change the number of queries that are concurrently
executed at the local DBMS (multi-programming
degree).
The QoS-OES (Costa & Furtado, 2008)
scheduler is an example of QoS-oriented query
scheduler that can be used in such context. Such
module is a generic external scheduler that is
used as a middleware between the DBMS and its
users. The QoS-OES is capable to estimate query
execution time in a multi-query environment and
to commit itself to execute a submitted query
by a certain deadline, as soon as the specified
deadline time is greater than the predicted query
execution time.

nodes of a distributed environment is NP-hard
(Loukopoulos & Ahmad, 2000), some heuristics
should be used. The heuristics used in this Sec-
tion are QoS-oriented, which means they intend
to increase the system's SLO-achievement rate
(SLO-AR).
The SLO-AR (Costa & Furtado, 2008b) is
a performance metric that aims to indicate how
well a system is performing on executing jobs by
specified service level objectives. It is defined
as the relation between the number of queries
whose execution finishes by the required deadline
(N) and the number of queries in the submitted
workload (W).
Therefore, the SLO-Oriented replica selection
and placement strategy aims at increasing the
number of queries that the system executes by
the specified SLOs. In order to do that, a benefit-
based mechanism is implemented by a Replication
Manager (RM), as it is described below.

INTER-SITE QOS-ORIENTED
DYNAMIC REPLICATION

Caching and Replication for QoS The RM monitors the number of times that a

SLO-objective cannot be achieved due to the
inexistence of a certain dataset (e.g. facts table
fragment) at a certain site and computes the total
benefit value ( β ) that the system would have if
a replica of such dataset is created. When β is
greater than a threshold value, the system considers
the possibility of creating the data replica at the

Grids are highly heterogeneous environments
in which the use of database replicas can lead
to great performance improvement and to high
QoS levels. But as the problem of choosing the
optimal number of database replicas and doing
the optimal placement of such replicas into the