FW: "Performance Counters"

Not sure why this didn't go to the list.  

From: Mark W. Farnham [mailto:mwf_at_rsiz.com] Sent: Friday, April 27, 2007 12:27 PM
To: 'jeffthomas24_at_gmail.com'
Subject: RE: "Performance Counters"  

"if any" is on target. The type of metrics noted are the rough equivalent of reporting engine RPMs without knowing the gear ratio - or even whether the tires are touching the ground, and no idea of the fuel consumption rate or whether the vehicle was moved in the correct direction or at all.  

Sensible metrics relate to service times to perform critical functional operations, usually expressed as something like: "We averaged half the allowable service time and never exceeded 85% of the allowable service time for processing an order," or "we successfully captured 99.7% of the data with an experimental design allowing for a minimum datapoint capture rate of 95%.  

Truly useful metrics then relate to the cost of service for remaining within tolerated variance from pre-determined service levels so that a determination of whether more or less money should be spent on the system can be made considering the value of achieving the stated service levels.  

Wasting time and money concocting useless metrics so there is a correspondence with existing useless metrics from another system is like watering a noxious weed during a drought.  

Regards,  

Mark W. Farnham  

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Jeffery Thomas
Sent: Friday, April 27, 2007 11:02 AM
To: Oracle-L Freelists
Subject: "Performance Counters"

<snip>

The ones identified for SQL Server were:

SQL Server: SQL Statistics: SQL Compilations/Sec counter must not exceed 100 per second over a 2 minute period.
Average Disk Queue Length, averaged over a 2 minute period must not exceed 2.5
SQL Server: Buffer Cache Hit Ratio should exceed 90% measured over a 2 minute period.
The CPU usage must not exceed 35% averaged over a 2 minute period. The SQLServer:LockWaitTime must not exceed 2ms measured over a 2 minute period

If you've got a vendor building an application for you -- what kind of benchmarks -- if any -- actually make sense with respect to performance criteria that the vendor must meet?
<snip>

--
http://www.freelists.org/webpage/oracle-l