Design of Algorithms for Big Data Analytics

Processing of high volume and high velocity datasets requires design of algorithms that can exploit the availability of multiple servers configured for asynchronous and simultaneous processing of smaller chunks of large datasets. The Map-Reduce paradigm provides a very effective mechanism for designing efficient algorithms for processing high volume datasets. Sometimes a simple adaptation of a sequential solution of a problem to design Map-Reduce algorithms doesn't draw the full potential of the paradigm. A completely new rethink of the solution from the perspective of the powers of Map-Reduce paradigm can provide very large gains. We present here an example to show that the simple adaptation does not perform as well as a completely new Map-Reduce compatible solution. We do this using the problem of finding all formal concepts from a binary dataset. The problem of handling very high volume data is another important problem and requires newer thinking when designing solutions. We present here an example of the design of a model learning solution from a very high volume monitoring data from a manufacturing environment.