|
|
Archives of the TeradataForum
Message Posted: Wed, 10 Sep 2003 @ 13:24:24 GMT
Subj: | | Re: CPU vs Disk IO |
|
From: | | Judge, James A |
There is an indicator for "run-away" queries (probably more accurate description are either "product join" or "duplicate row check"
steps). This has been attributed to long time performance expert Larry Higa, and thus called the LHR (Larry Higa Ratio). I have found it an
accurate indicator of a problem query. The formula is (CPUTime * 1000)/DiskIO . This gives you milliseconds of CPU per IO. In normal
operations this ratio is probably around .01 to 1 ms per IO. When you see it reach 20 and greater then you most likely have a query that is
seriously stressing the configuration. The product join or duplicate row check (during an INSERT/UPDATE operation obviously) will have been
running for minutes (as in GT than 10 or so) to reach this ratio I think (maybe it's just my refresh rates though). There was a web-
roundtable (WBT) session on the Teradata Education Network last year by Larry on this subject.
| |