Archives of the TeradataForum

Message Posted: Mon, 14 Jun 2004 @ 19:50:54 GMT

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

As John mentioned, there is an internal field in DBSControl that is used by DBS start-up to determine the number of AMP worker tasks that will be started for each AMP in the system. The maximum value is 120, but the default is 80. 80 has proven to be a good number over the years for most systems, and a good point of balance between productive usage of the platform and loss of throughput caused by unexpectedly high demand. If you feel this number is too low for you, you would want to discuss this with your local account team, and probably work with the Support Center to get their support in implementing a change. Because this is a fairly basic (and somewhat delicate) function within Teradata, and can have reverberations on performance if changes are not well thought out, you will need to have some conversations beforehand, and maybe it will end up not being in your best interest after all.

On the downside, each AWT engaged in active work requires on average between 1 and 2 MB of memory, for each AMP. If you have 10 AMPs per node this can add up. This is less memory for other things, such as the FSG cache.

The theory behind setting a limit is that it is better for overall throughput to put the brakes on before exhaustion of all resources is reached. In addition, by keeping all AMPs to a reasonable usage level, this increases parallel efficiency.

One advantage of using AWTs as a sort of governer, rather than sessions, is that messages that cannot get an AWT will queue up automatically by type (aborts, for example, line up before new work) and by priority scheduler priority. That is not the case with session-handling.

Thanks, --Carrie

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