Archives of the TeradataForum
Message Posted: Fri, 27 Aug 2004 @ 17:23:00 GMT
<-- Anonymously Posted: Friday, August 27, 2004 13:05 -->
Searching the T-data archives for "flow control", I've found it described as a response to running out of available amp worker tasks, for example "If your system ran out of available AWT's it would go into 'flow control' stopping any new work from entering until AWT's became available."
I was thinking that running out of amp worker tasks and going into flow control were two different symptoms, the first being a shortage of available slots to handle work (60 awt per amp), and the second indicating bynet message congestion on one or more nodes (msg count = #nodes + 5).
Let's say that a system constantly congests on amp worker tasks and is almost always in flow control, but CPU, I/O, & memory aren't maxed out. And let's say that queries are perfectly tuned and mostly primary index retrievals (an imaginary world). If the number of AWTs were bumped up (with NCR's blessing, of course), and you didn't congest on AWTs anymore, then would you expect flow control to ease up as well, or that it would get worse due to the increase in message traffic caused by running more work at once?
|Copyright 2016 - All Rights Reserved|
|Last Modified: 15 Jun 2023|