Sorry for hijacking the thread, but i have an answer for that.
IMHO having the computation application and the data management on different domains incurs a very high data transfer cost.
The traditional thinking of shipping the data from the DB to somewhere else (application code) to do the computation and then shipping the result back to the DB is putting an enormous overhead on data processing tasks that need to do this back and forth shipping many times. Even when using SQLite that lives on the same process as the computation application, this overhead is *very* considerable.
By moving the computation closer/inside the DB engine (with UDFs), this overhead vanishes. I have already seen many cases of people implementing workflows the traditional way (DB server, computation on the app. server) on high powered machines, being several times slower than me doing the same processing using madIS [*] on my low powered desktop.
All of the above apply for OLAP processing. For OLTP, the traditional approach is good enough.
lefteris. [*] http://code.google.com/p/madis/ On 12/10/12 21:41, Igor Tandetnik wrote:
... In light of this, I don't quite see what you expect to gain from turning it into a UDF, as opposed to simply loading the sequence from the database into memory with a vanilla SELECT statement, and working on it in your application code.
_______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
