Archives of the TeradataForum

Message Posted: Wed, 02 Apr 2008 @ 23:03:25 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Hi Bill,

I have no idea whether or not you can adjust the keep alive timer. In fact I don't even know if there is one - and if there is, I would have guessed it is much longer than 1 second as it can take several minutes before TD recognises orphaned sessions and cleans them up.

However, I did want to suggest another possibility. In my experience many 3rd parties don't really understand how MPP systems work - particularly in tasks like this where you are copying rows from one table to another.

In a significant proportion of cases they will run a query on the source system and row by row load the result into TD. The problem with this is that TD is now a MUP - Massively Unparallel System. Only one AMP is doing any work at any particular time. Translation - performance sucks profusely!

The key to getting high data load rates - and you can get quite high rates with JDBC is to get the other AMP's working at the same time. This is achieved with TPump style techniques and includes using multiple sessions, leveraging query caching and sending multiple data parcels per request. All of these can be achieved using JDBC and result in much higher throughput (without tinkering with the KEEP ALIVE POLL setting - assuming there is on and it can be tinkered with).

Having said that, have you considered using TPUMP or TPT? If you need some complex transformations that can't be done in the SQL or post load, you could certainly approach TPUMP type performance levels using the techniques I mentioned above.

All the best

Glenn Mc

	<Prev	Next>		<<First	<Prev	Next>	Last>>