QoS-Oriented Grid-Enabled Data Warehouses - Data Warehousing Design and Advanced Engineering Applications - page 159

Figure 3. Global physical allocation example

query scheduling is transforming the submitted
query (job) into other ones (tasks) that access
the physically distributed global relations. Such
transformation is similar to the ones presented at
(Furtado, 2004b).
The generated tasks are assigned to sites by
a Community Scheduler. At each site, a Local
Scheduler is responsible to manage task execu-
tion. If domain specific physical layout is used,
the Local Scheduler must also do the conversion
of the globally specified task into other queries
that access the physically existent relations. The
Community Scheduler specifies the necessary
requirements (execution deadlines) of each task
(rewritten query) in order to execute the user's
query by the desired QoS level. Then, there is a
task execution negotiation phase among the Com-
munity Scheduler and Local Schedulers, in order
to verify if each of the rewritten queries (tasks)
can be executed by its deadline. If any task can-
not be executed by the specified deadline, then
user is notified that the required SLO cannot be
achieved and a new one must be specified, or the
query execution would be canceled.

The Community Scheduler assigns que-
ries to sites considering the Global Physical
Model. Hence, if local domain uses some data
partitioning policy different from the one used
at the global level (e.g. large dimension table
partitioning, as in the abovementioned NPDW
strategy), then the Local Scheduler should
transform the globally specified query into the
ones that should be locally executed. Besides
that, results merging should also be done at local
site level in order to send back a single result,
which is correspondent to the query (task) the
site has received.

EVALUATING IF A SLO
CAN BE ACHIEVED

Let's consider a user-submitted query Q with a
deadline interval d . The system must estimate
the query's total execution time ( tet ) and com-
pare it with d in order to verify if the query can
or cannot be executed by its deadline. When the
system estimates that the execution can be done