Monday, February 21, 2011


-- By RK

Database Indexing for the Lookup SQL - A myth

I have seen people suggesting an index to improve the performance of any SQL. This suggestion is incorrect - many times. Specially when talking about indexing the condition port columns of Lookup SQL, it is far more "incorrect".

Before explaining why it is incorrect, I would try to detail the functionality of Lookup. To explain the stuff with an example, we take the usual HR schema EMP table. I have EMPNO, ENAME, SALARY as columns in EMP table.

Let us say, there is a lookup in ETL mapping that checks for a particular EMPNO and returns ENAME and SALARY from the Lookup. Now, the output ports for the Lookup are "ENAME" and "SALARY". The condition port is "EMPNO". Imagine that you are facing performance problems with this Lookup and one of the suggestion was to index the condition port.

As suggested (incorrectly) you create an index on EMPNO column in the underlying database table. Practically, the SQL the lookup executes is going to be this:


The data resulted from this query is stored in the Lookup cache and then, each record from the source is looked up against this cache. So, the checking against the condition port column is done in the Informatica Lookup cache and "not in the database". So any index created in the database has no effect for this.

You may be wondering if we can replicate the same indexing here in Lookup Cache. You don't have to worry about it. PowerCenter create "index" cache and "data" cache for the Lookup. In this case, condition port data - "EMPNO" is indexed and hashed in "index" cache and the rest along with EMPNO is found in "data" cache.

I hope now you understand why indexing condition port columns doesn't increase performance.

Having said that, I want to take you to a different kind of lookup, where you would've disabled the caching. In this kind of Lookup, there is no cache. Everytime a row is sent into lookup, the SQL is executed against database. In this scenario, the database index "may" work. But, if the performance of the lookup is a problem, then "cache-less" lookup itself may be a problem.

I would go for cache-less lookup if my source data records is less than the number of records in my lookup table. In this case ONLY, indexing the condition ports will work. Everywhere else, it is just a mere chanse of luck, that makes the database pick up index.

I'm sure you will have many questions regarding Lookups after reading this blog article. I'm ready to answer, fire away.

No comments:

Post a Comment