Database Reference
In-Depth Information
Amazon S3
A similar measurement study was conducted on Amazon Simple Storage Service
(S3) for 11 days. In S3, storage consists of objects within buckets, so our writer
updates an object in a bucket with the current timestamp as its new value, and
each reader reads the object. In this experiment, measurements for the same five
configurations as SimpleDB's case are conduced, including a writer and a reader
run in a single thread, different threads, different processes, different instances,
and different regions. Amazon S3 supports two types of write operations, namely
standard and reduced redundancy. A standard write operation stores an object so that
its probability of durability is at least 99:999;999;999%, while a reduced redundancy
write aims to provide at least 99:99% probability of durability. The same set of
measurements was performed with both standard write and reduced redundancy
write.
Documentation states that Amazon S3 buckets provide eventual consistency for
overwrite
PUTS
operations. However, no stale data was ever observed in this study
regardless of write redundancy options. It seems that staleness and inconsistency
might be visible to a customer of Amazon S3 only in executions in the event of a
failure in the particular nodes of the platform where the data is stored, during the
time of their access; this is a very low probability event.
Microsoft Windows Azure Table Storage and Blob Storage
The experiment was also conducted on Microsoft Windows Azure Table Storage
and Blob Storages for 8 days. Since it is not possible to start more than one process
on a single instance, specifically for a
web role
in this experiment, measurements
for four configurations are conducted: a write and a reader run in a single thread,
different threads, different instances or different regions. On Azure Table Storage a
writer updates a property of a table and a reader reads the same property. On Azure
Blob Storage a write updates a blob and a reader reads it.
The measurement study observed no stale data at all. It is known that all types of
Microsoft Windows Azure Storages support strong data consistency [
165
] and this
experiment confirms it.
Google App Engine Datastore
Similar to Amazon SimpleDB, Google App Engine Datastore keeps key-accessed
entities with properties and it offers two options for reading:
strong consistent read
and eventual consistent read. However, the observed behavior for eventual consistent
read in the Datastore is completely different from that of Amazon SimpleDB.
