“Search engines primitive”: IBM Master Inventor

He is known as the leading innovator of transaction management. He is also the recipient of several awards for his innovative contributions to the development of database systems. In the last 24 years in IBM's Almaden Research Center, Dr C Mohan has collected 34 patents on various inventions and, in 1997, was named the IBM Master Inventor. The same year he was named an IBM Fellow, IBM's highest technical position.

On his visit to IBM's Research Labs in India, Dr C Mohan spoke to Sam Varghese of CIOL on subjects ranging from the complexity in managing unstructured data to making query processing faster and more accurate. Excerpts from the interview:

What brings you to India?

Among other things, we want to give the employees of India Research Labs a feel for the technical side of the game. Research workers across the world, and not just in India, try to compensate for the career ladder by switching to the management side. We want to obviate this wandering and give those inclined to research an impetus to continue in their area of competence.

What kind of research work gets done out of IBM's India Research Labs?

We are attempting to analyze transcripts of call center conversations in a bid to enable business intelligence operations. We are working with Daksh to facilitate a better understanding of the unstructured data generated during such conversation to be able to put a finger to request compliance and several other such spin-offs.

How would you describe the state of web searches as it is today?

Web search engines are as primitive as they could be. They fumble at the first hint of a fancy search query. When we think of keywords as values for specific fields, we are looking at a clarity with regard to the semantics. Our association with Daksh is connected to this.

Today, it's more about work on structured data and data being circulated via links and page rankings when it comes to a search. When we started work on what we call the Omnifind programme, we were looking at an enterprise kind of search — more relevant in the intranet context. We needed to think beyond the gamut of links.

Where would you put DB2 with respect to its capability to search for both structured and unstructured data?

Now, we are working on enhancing DB2 to support XML in a more complex way. We'll be handling, storing and indexing XML in a 'first class' way. Speaking of DB2, it boasts a 'health monitor', a salient means of ensuring system health.

We can store XML, marking a certain node as the child of another. The whole process can be viewed as some sort of 'hierarchy instrument' and makes life that much more easier for the administrator in terms of the shredding and database design.

We are trying to exploit such inferred information to help process queries and are working on taking functionality to the next level from database commoditisation. We want to lend a new meaning to DBMS.

Can you shed some light on the work going on in the area of query processing?

Reoptimisation of queries is another area we are exploring avidly. We try to work on dynamically changing query execution than wait for the original query plan to get completed. We now have algorithms that can come handy while handling queries that involve large amounts of data that may share implicit relationships while not being totally explicit.

Despite the highly structured queries, the query optimizer gets about picking the best way of ferreting the relevant information. Adaptive query processing, with the query optimizer enabling such flexible query execution, can even facilitate a match-up between the optimizer and the dynamic execution results.