Performance Counter - % Disk Time and Avg. Disk Queue Length

Question

% Disk Time - How often is the disk busy? If it's busy almost all of the time, and there is a large queue, the disk might be a bottleneck. This value is displayed as a percentage and is capped at 100%.

This is the definition from Microsoft. However, as I saw from the % Disk Time data of my machine's performance monitor, it was always around 300 and 400. Why this does not match the definition of % Disk Time?

Avg. Disk Queue Length - How often is disk busy? If it's busy almost all of the time, and there is a large queue, the disk might be a bottleneck. This counter displays %Disk Time as a decimal with no defined maximum. (A %Disk Time of 100% equals an Avg. Disk Queue Length of 1.0.) This counter is recommended for disk sets where the combined activity of the disks can exceed 100% of a single disk.

In addition, I heard someone said that "If the Avg. Disk Queue Length is larger than the number of CPU cores of the machine then it will cause problem" I don't know if this is true or not because I cannot find anything about this on the Internet.

Answer 1

% Disk Time - Logical disk or Physical disk?

From Microsoft: http://support.microsoft.com/kb/310067/en-us?p=1

This behavior can occur because some controllers allow the operating system to use overlapping input/output operations for multiple outstanding requests. The disk performance counters time the responses by using a 100 nanosecond precision counter, and then report the cumulative statistics for a given sample time. This sample time could go over 100 percent if, for example, you have 10 requests that completed in 2 milliseconds each in a 10 millisecond sampling interval. If you have multiple disks in a Raid arrangement, the overlapped input/output happens because the operating system can read and write to multiple disks, and this could show values that are higher than 100 percent for this counter.

You using RAID?

Avg. Disk Queue Length - Correlating queue length to the number of processor cores you have sounds rather dubious. I've never heard anyone offer that advice before. Unless you are experiencing major CPU load as well, your CPU can handle disk operations way faster than any disk I know of could perform them, which is why they stack up in the form of disk queue length in the first place.

Queue length is the number of outstanding operations. You want to keep it as low as possible. Any amount of queue length means that there are pending disk operations that have not been serviced yet. Ideally you don't want any of that. Realistically, because disks are slow, you'll always see some, but you want to keep it as low as possible.

Again, it depends on if you're measuring at the Logical disk level or the Physical disk level. A logical disk can consist of many physical disks. If the queue length of the logical disk is higher than the number of physical disks/spindles that make up that logical disk, then you're probably running into a bottleneck situation.