Thanks a lot for http://ids.snu.ac.kr/w/images/c/cf/ICUIMC2012-Myung.pdf 

Advertisements

1. Matrix multiplication. A lot of real life problems can be solved via a linear algebra approach. Among the many parallel  algorithms for matrix multiplication, Fox’s Algorithm and Cannon’s Algorithm perform the best. With column/row decomposition strategy we are able to process such approaches via map-reduce.

2. After translating a real life problem into matrices we are able to cluster the related data. I.e. Kmeans clustering is a well-known clustering algorithm aiming to cluster a set of data points to a predefined number of clusters. More details of this analysis can be found in this paper

3. Beside the clustering of unstructured data we have to form groups. More details of this analysis can be found in this paper.

4. Counting. This is a simple MapReduce computation that is often use to explain the MapReduce programming model. Very very basic but important. 

5. Filtering. As basic as counting but most time not considered. Drop out unnecessary data.

 

For all join-operations use iterative map-reduce on temporary tables 😉

while (condition)

{

          MapReduce()

          UpdateData()

}

 

 

 

 

When discovering your business you get a 360 degree. As a result you will take corrective actions, Like drop out the legacy system A, B, … and replace them with C. A more effective approach is to analyse the business based on the available data and business rules. From these two “things” you get valid information. But, only for already established business – still a gap exists :(. The processes and their limits are the missing link. If a department don’t know what the limits are, be worried ;-).

 

Dear reader, please give me feedback.

While reading the following article http://www.linuxforu.com/2011/02/up-close-and-personal-with-nosql/ i found some interesting aspects regarding “Triple stores”, because the most business-cases play with subject-predicate-object semantic. 

The technical approaches of CEP are part of the NextBI “speed layer”. CEP handles a high amount of simultane occuring events. The whole spectrum of occuring events are flatted and a decision will be extracted (in “near” realtime). The extraction of a decision is equal to that what the real time approaches of NextBI intend. Take multiple channels of information, combine it with specific context (like current geographical location, weather and so on) and find a best matching solution. Like the intelligent search of web search engines. Take the information of all in the internet web sites (typically batch jobs) and bring it together with real time information.  

Consider a stable connection between server and client. If so, the semantic of SOA approaches will change. For example with WebSocket. A established WebSocket allows the service to broadcast – without any polling by the clients.