Database Reference
In-Depth Information
Consistent Read
Eventual Consistent Read
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0
200
400
600
800
1000
Time elapsed from completing write until starting read (ms)
Fig. 5.2
Probability of reading freshest value
2. A writer and a reader run in different processes but in the same instance that is
also in the same geographic domain as the data storage in
us-west
region. In this
case, read and write requests still have the same IP address.
3. A writer and a reader run on different instances but both are still in the same
region. In this case, requests originate from different IP addresses but from the
same geographical region.
4. A writer and a reader run on different instances and different regions, one in
us-west
region and one in
eu-west
region. In this case, requests originate from
different IP addresses in different regions.
The measurement is executed once every hour for 11 days from October 21,
2010. In total 26;500 writes and 3;975;000 reads were performed for accessing
from a single thread. Since only one thread is used in the first study, the average
throughput of reading and writing are 39:52 per second and 0:26 per second,
respectively, where each measurement runs at least for 5 min. The same set of
measurements was performed with eventual consistent read and with consistent
read.
In the study of accessing from multiple threads and processes, each experiment
was run for 11 days as well. In all four cases the probability of reading updated
values shows a similar distribution as in Fig.
5.2
. Therefore, it is concluded that
customers of Amazon SimpleDB see the same data consistency model regardless
of where and how clients are placed. Hence, this section will focus on reporting
observations of accessing from single thread with regards to two consistency
options, read-your-write consistency and monotonic read consistency respectively.
