 |
 |
Archives of the TeradataForum
Message Posted: Mon, 24 Jun 2002 @ 19:16:38 GMT
| Subj: | | Re: TDQM and Limits of CPU |
| |
| From: | | Thomas F. Stanek |
Al,
There are several ways to accomplish what you want to happen. The least sophisticated is to use the milestone limits in PSF. You could
set up performance groups with different weights and policies and have queries switch from one to another as they reach certain CPU
thresholds. Keep in mind that this is at the session level. I'm not sure if MicroStrategies establishes new sessions with each request or
not. If it does, this will work fine. If not, you will need to resort to a PM/API based application (i.e. either Teradata Manager or a
custom written application). Either way, you would have to define the criteria by which you would like to reduce the priority for a given
query. The nice thing about a custom application is that you can establish whatever rules you like based on a variety of variables. PM/API
provides a wealth of details about a given query. FYI, Kiki Sanchez of Continental Airlines gave a very interesting presentation at
Partners where he outlined the PM/API application they developed and he did a live demonstration of the tool in action.
I'd have to disagree that PSF is more "art than science". It seems to provide fairly predictable results, considering the range of
possible workload combinations and the fact that it has no influence over several factors such as locking contention, some system work, IO,
etc. It seems to do pretty much what it is designed to do, which is to dynamically assign weights to active tasks in an effort to allocate
the CPU resources accordingly. Probably the biggest weakness is the lack of detailed documentation and education about the capabilities of
the tool. I know that NCR has worked hard to fix that with presentations at the regional User Groups as well as Partners.
By the way, I have heard that there is a tool called Proxima that has a Pm/API interface that can provide the same functionality as the
custom application approach mentioned above. Does anyone have any hands on experience using it?
Thomas F. Stanek
TFS Consulting
www.tfsconsulting.com
| |
