CloudDB AutoAdmin - Large Scale and Big Data: Processing and Management - page 376

11.6.2 e XPerimental e valuation
We implemented two sets of experiments to evaluate the effectiveness of our adap-
tive replication controller in terms of its effect on the end-to-end system throughput
and the replication delay for the underlying database replicas. Figure 11.8 illustrates
the setup of our experiments using Amazon EC2 platform. In the first set of experi-
ments, we fix the value of the tolerance component ( delay tolerance ) of the SLA replica-
tion delay to 1000 milliseconds and vary the monitor interval ( intvl mon ) among the
following set of values: 60, 120, 240, and 480 seconds. In the second set of experi-
ments, we fix the monitor interval ( intvl mon ) to 120 seconds and adjust the SLA of
replication delay ( delay sla ) by varying the tolerance component of the replication
delay ( delay tolerance ) among the following set of values: 500, 1000, 2000, and 4000
milliseconds. We have been evaluating the round-trip component ( delay rtt ) of the
replication delays SLA ( delay sla ) for the database replicas in the three geographical
regions of our deployment by running ping command every second for a 10-minute
period. The resulting average three round-trip times ( delay rtt ) are 30, 130, and 200
milliseconds for the master to slaves in us-west , us-east , and eu-west , respectively.
Every experiment is executed for a period of 3000 seconds with a starting workload
of 220 concurrent users and database requests with a read/write ratio of 80/20. The

Cloudstone benchmark

M/N satisfies pre-defined read/write ratio

e controller
Monitor and
manage to scale

Replication within and across regions

Master

MySQL Proxy

M/N read
write split

Slave
us-west- 1

Slave
us-west- 2

Slave
us-east- 1

Slave
us-east- 2

Slave
eu-west- 1

Slave
eu-west- 2

FIGURE 11.8 Adaptive replication controller.