Recently I was working on an integration project where the database used was MS SQL server and the integration platform was Oracle SOA Suite.
The service itself was quite simple – to fetch some data from SQL server and Enqueue it on JMS queue. We used DB Adapter for polling the database. We used the Delete strategy and the service was distributed on a cluster.
Once the ids were enqueued, there was a seperate EJB-based webservice which queried same database to create the canonical. We have used JPA Entity Beans for ORM. There is a particular query to get some extra information from a table which does not have foreign-key relation. The query used a single parameter – a string.
However we observed a huge performance issue for SQL server as well as the website hosted on the database. We observed 99% CPU usage.
It was our SQL DBA who found out the issue. The column in database was varchar, the index was based on same. However, the query parameter that got sent to the database was using nvarchar. This caused a full table scan and completely skipped the index.
The solution use “sendStringParametersAsUnicode” property of Weblogic SQL Server driver. By default everything gets sent as Unicode from JDBC Driver, using “sendStringParametersAsUnicode=false”, we made the driver send the parameter as varchar and immediately saw the difference. CPU usage was down to 1%.
This underscores the point that Frameworks and Engines abstract out a lot of features, but it is necessary to understand you database to make optimal use of it.
Reference – http://msdn.microsoft.com/en-us/library/ms378988.aspx