Database Reference
In-Depth Information
are maintained by Network Time Protocol (NTP)
4
on Amazon EC2. The system
clock is set to synchronize with multiple time servers every second to have a better
resolution. More details in dealing with the clock synchronization issue in the cloud
will be discussed in Sect.
6.1
.
With the customized Cloudstone
5
and the heartbeat plug-in, it is possible
to achieve the goal of measuring the end-to-end database throughput and the
replication delay. In particular, two configurations of the read/write ratios, 50/50
and 80/20 are defined. More over, three configurations of the geographical locations
based on
availability zones
and
regions
are also defined and listed as follows where
availability zones are defined as distinct locations within a region and zones are
separated into geographic areas or countries:
Same zone
: all slaves are deployed in the same availability zone of a region of
the master database.
Different zones
: all slaves are in the same region as the master database, but in
different availability zones.
Different regions
: all slaves are geographically distributed in a different region
from where the master database is located.
The workload and the number of virtualized database replica servers start with a
small number and gradually increase at a fixed step. Both numbers stop increasing
if there are no throughput gained.
Clock Synchronization in Cloud
The clock synchronization issue refers to the fact that internal clocks of physical
machines may differ due to the initial clock setting and subsequent clock drift.
It results in time differences between two machines even though both machines
perform the read operation at the same time. This issue could also happen to
instances in the cloud environment, if two instances are deployed in distinct physical
machines where the clock is not shared. As a matter of fact, it has been observed
by [
199
] that all instances launched by a single Amazon EC2 account never run in
the same physical node. Hence, all running instances that belong to a single account
will exhibit the clock synchronization issue.
The replication delay in experiments is measured based on committed local
timestamps on two or more virtualized database replica servers. Thus, the clock
synchronization issue also exists in the replication delay. As the study is more
interested in the changes of replication delay, rather than that of accuracy, an
average relative replication delay is adopted to eliminate the time differences
introduced by the clock synchronization issue. The average relative replication
