Archives of the TeradataForum

Message Posted: Tue, 25 Jul 2006 @ 17:56:32 GMT

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

In my role I review many machines that run at 100%. The question is if the machine runs at 100% almost 100% of the time. When that is the case one really needs to use workload management to keep AMPWorkerTasks under control. Specifically limit long duration all-amp queries to 10 or less. Keeping the number low (at less than 10) helps, as long as the long-duration queries get the system running at close to 100%. Adding more that drives the system beyond 100% does not help anything and hurts small bursts from tactical queries. Coexistence systems benefit more from workload management as they do not run beyond 100% as well as a single generation system. Still the coexistence systems run well at 100%, so long as the slow nodes are running at 100% and are not getting overbooked with work.

Your experience will vary. Know that opinions are all over the place on your question. Here are my opinions.

The following are guidelines I use to advise customers:

- when running applications that require human dialogs, web apps, customer interaction queries then running a machine at 60% of CPU capacity or lower is good. Between 60% and 70% is yellow. In general I find over 70% to be troublesome. This measure is over a 24 hour period. I have one customer running at slightly over 40% and is looking to upgrade their system. The phrase "your experience will vary" applies.

- when running less response time intensive applications running at 70% to 85% seems OK. Again this is measured over a 24 hour period.

- when running batch and "who cares about response time" then running below 95% seems to be OK.

At some point in time, usage grows to the point where you no longer have time to rerun failed loader jobs or recover from a down node. This is the point-in-time when life gets tense.

Usual straight-line system growth ranges from 1.5% to 4% or more per month. One needs to observe, plot and report this growth monthly so that the budgeting process can predict spending. Otherwise unhappy users are in your future.

Best wishes, your system is getting close.

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