CloudDB AutoAdmin - Large Scale and Big Data: Processing and Management - page 385

TABLE 11.2
Number of Provisioned Database Replicas

Workload/Rule R1 R2 R3 R4

80/20 4 3 5 5

50/50 5 4 7 6

In these figures, the X - axis represents the elapsed time of the experiment while the
Y - axis represents the SLA satisfaction of the application workload according to
the different elasticity rules. In general, we see that, even for this relatively small
deployment, the incorporation of SLA-based rules can show improved overall SLA
satisfaction of different workloads of the application. The results show that the SLA-
based rules ( R3 and R4 ) are, by design, more sensitive for achieving the SLA satis-
faction, and thus they react earlier than the resourcebased rules. The resource-based
rules ( R1 and R2 ) can accept a longer period of SLA violations before taking any
necessary action (CPU utilization reaches the defined limit). The benefits of SLA-
based rules become clear with the workload increase (increasing the number of users
during the experiment time). The gap between the resource- and SLA-based rules
is smaller for the workload with the higher write ratio (50/50) due to the higher
contention of CPU resources for the write operations and thus the conditions of the
resource-based rules can be satisfied earlier.
Table 11.2 shows the total number of provisioned database replicas using the differ-
ent elasticity rules for the two different workloads. Clearly, while the SLA-based rules
achieves better SLA satisfaction, they may also provision more database replicas. This
trade-off shows that there is no clear winner between the two approaches, and we can-
not favor one approach over the other. However, the declarative SLA-based approach
empowers the cloud consumer with a more convenient and flexible mechanism for
controlling and achieving their policies in dynamic environments such as the cloud.

11.8 RELATED WORK
Several approaches have been proposed for dynamic provisioning of computing
resources based on their effective utilization [7,12,23]. These approaches are mainly
geared toward the perspective of cloud providers. Wood et al. [23] have presented an
approach for dynamic provisioning of virtual machines. They define a unique met-
ric based on the data consumption of the three physical computing resources: CPU,
network, and memory to make the provisioning decision. Padala et al. [12] carried
out black box profiling of the applications and built an approximated model, which
relates performance attributes such as the response time to the fraction of proces-
sor allocated to the virtual machine on which the application is running. Dolly [5]
is a virtual machine cloning technique to spawn database replicas and provisioning
shared-nothing replicated databases in the cloud. The technique proposes database
provisioning cost models to adapt the provisioning policy to the low-level cloud
resources according to the application requirements. Rogers et al. [15] proposed two