Cosmos is Microsoft's internal BigData analysis platform.
Cosmos is composed of a distributed compute component (somewhat comparable to Hadoop's Map-Reduce, using the Microsoft Dryad solution, which (unlike Map-Reduce) allows an arbitrary DAG of computation. Cosmos supports a SQL-like syntax (similarly to HIVE/PIG) and includes a distributed storage component (comparable to HDFS); Overall, Cosmos provides highly scalable, reliable, fault-tolerant and automatically scaled compute operations on huge data sets.
According to , , Cosmos allows the use of SQL-like syntax such as
source = EXTRACT col1, col2 FROM “A”
Data = SELECT * FROM source where (condition)
to extract data without the need for explicit Map/Reduce primitive use by the user-developer.
The SCOPE language, just like SQL, also supports  operators such as where, join, reduce, as well as user-defined operators. These include reducers (basically, input parsing) and other programmable operations in user code [1,3], and generates a parallel, optimized  'execution plan' for the defined queries.
Cosmos is used within Microsoft extensively, across a huge number of servers , stores a large amount of data, and processes a vast amount of data daily: "Every day we process hundreds of petabytes of data from Bing, AdCenter, MSN, and Windows Live" [Software Development Engineer, Principal-BING Job] .