Saturday, March 8, 2008

ITM 6.1 FP01 changes the Physical Disk/ % Disk Time attribute

ITM 6.1 FP01 changes the way %Disk Time is derived by overlaying it with a new calculation (i.e., 100-%Idle Time). In APAR IY82459 IBM describes the original %Disk Time as a Perfmon miscalculation because it renders percentages greater than 100%.

According to Microsoft's Article ID 310067 this behavior is by design. Actually, as strange as it seems having a percentage attribute in %Disk Time greater than 100% is not a bug. You can check out these resources describing why:;en-us;310067

Microsoft's recommendation in Article ID 310067 is not to obfuscate the original % Disk Time but to derive a new one based on the newer %Idle Time counter to create a new disk usage (busy) time attribute or just use the % Idle Time counter by itself. Further more, Microsoft also provided a newer counter called Average Disk Queue Length to be used instead of % Disk Time (this is described in the previous mentioned Blog entry).

Here in lies the problem. Monitoring applications like ITM 5.x and ITM 6.1 were first developed in the mid to late 90's (yes ITM6 was originally Omegamon) and neither have been updated to provide many of the updated Perfmon object counters. Take for example the Perfmon Physical Disk performance object. On an XP/2000 system there are 21 performance counters for the Physical Disk performance object. The ITM 6.1 NT_PhysicalDisk attribute group only includes 12 attributes. In fact, ITM 6.1 NT Phyiscal Disk doesn't even include the two Microsoft recommended counters, %Idle Time and Average Disk Queue Length.

The other issue is that in ITM 6.1 the % Disk Time value is defined as a RANGE: 0-100 attribute and situations can not been written to handle values greater than 100. It appears that the quick fix in ITM 6.1 FP01 was to recalculate the value based on Microsoft's recommendation and maybe that is the best fit solution.

However, IMHO I think what IBM should have done was take the RANGE qualifier off the % Disk Time and left the original %Disk Time counter value alone. Then they should have added the newer Perfmon counters to the NT Physical Disk attribute group (In fact they really should update all of the Performance Objects with the newer counters). Whith this approach the NT Physical Disk %Disk Time attribute would stay in its original format and user's could then take advantage of the newer Microsoft recommended performance counters (e.g., %Idle Time and Avg. Disk Queue Length).

John Willis

No comments: