Archives of the TeradataForum

Message Posted: Fri, 09 Sep 2005 @ 16:00:39 GMT

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

From: cbarrineau

You do the load balancing on UNIX for CLI connections by setting up the hosts file with one entry for each of the nodes. The utility picks them in a pseudo-random manner (really random or round-robin? I've gotten different answers) and spreads the connections around. Since the connections are an extremely small part of the UNIX overhead, this is usually Good Enough.

For ODBC (and I believe JDBC) connections, you do the same thing--there's a Teradata-specific extension to the ODBC standard that allows you to define more than one machine. See the manual for details. (From memory, I think it's DBConnectionX [extending DBConnection], where X ranges from 2 to the highest-numbered node. I don't know whether it'll parse correctly if you put leading zeroes in--never tried it.)

You can play games with these connections, and I know of larger sites that have done so.

There are also custom load balancing solutions for extremely large systems, but those are very special cases, and are very seldom needed.

All this is true _if_ your system is correctly configured. If you have all connections on a large system running to one node, you've got a bottleneck _and_ an unnecessary single point of failure in a mission-critical system.

Not that that stops people from doing it, or from getting lucky and not having that node fail before they change it.

Usually less grouchy in the morning,

John A

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