Database Reference
In-Depth Information
5.2
Staleness of Data on Different Cloud Platforms
Amazon SimpleDB
Amazon SimpleDB is a distributed key-value store offered by Amazon. Each key
has an associated collection of attributes, each with a value. For these experiments,
a data element is taken to be a particular attribute kept for a particular key, which
identifies, in SimpleDB terms, an
item
. SimpleDB supports a write operation call via
PutAttributes
and two types of read operations, distinguished by a parameter in the
call to
GetAttributes
:
eventual consistent read
and
consistent read
. The consistent
read is supposed to ensure that the value returned always comes from the most
recently completed write operation, while an eventually consistent read does not
give this guarantee. This study investigates how these differences appear to the
customers who consume data.
Amazon SimpleDB is currently operated in several independent geographic
regions and each of them offers a distinct URL as its access point. For example,
https://sdb.us-west-1.amazonaws.com
is the URL of SimpleDB operated in
us-west
region. It is used as the testbed in all experiments. The benchmark application for
Amazon SimpleDB is implemented in Java and runs in Amazon EC2. It accesses
SimpleDB through its REST interface. The writer writes timestamps, each of which
is 14 bytes of string data, in a key-value pair. The reader reads a value from the same
key-value pair using
eventual consistent read
or
consistent read
option. The study
of Amazon SimpleDB comprises of both parts based on the access patterns. The
access patterns determine the location options of EC2 instances that the writer and
the reader could reside, including options of being in the same region or in different
regions.
Access Patterns
In the first pattern, the writer and reader run in the same single thread on an
m1.small
instance provided by Amazon EC2 with Ubuntu 9.10. The instance is deployed
in the same region of SimpleDB, in the hope of minimizing the network latency.
Although, it is not guaranteed that data items from SimpleDB will be in the same
physical data center as the thread in EC2, using the same geographic region is the
best mechanism to the customer to reduce network latency. For this access pattern,
two consistency options,
read-your-write
and
monotonic read
are examined.
While in the second pattern, the writer and the reader are deliberately separated
to multiple threads, with the following configurations:
1. A writer and a reader run in different threads but in the same process. In this case,
read and write requests originate from the same IP address.
