COMPUTATIONAL RESEARCH in BOSTON and BEYOND (CRIBB)

As business and scientific datasets get larger and larger, they become increasing difficult to analyze in an efficient and productive manner. This is particularly true for analyses that are most conveniently expressed in terms of matrix and array computations.

In this talk, we describe Microsoft Cloud Numerics (http://www.microsoft.com/en-us/sqlazurelabs/labs/numerics.aspx), a distributed in-memory .NET library for performing numerical analysis on large dense and sparse datasets on Windows Azure. Data is exposed to the programmer via the notion of distributed multi-dimensional arrays that are supplemented by a large library of basic math, linear algebra, statistics and signal processing functions. The Cloud Numerics runtime then efficiently maps high-level operations on distributed arrays in .NET to highly-tuned lower-level primitives and library calls.

Although CloudNumerics is currently exposed only to .NET languages such as C#, F# and IronPython, the underlying distributed array infrastructure is designed to be bound easily and naturally to a number of other environments for data analysis. Towards the end of the talk, we will briefly cover some of the ongoing work on binding to environments other than .NET.