Is Disk I/O a Bottleneck?

Types of bottlenecks that can happen pertaining to disk can be due to two reasons –

a.       Pertaining to disk space

b.      Pertaining to time

The disk bottleneck pertaining to space can be identified easily by plotting following counters:

            Logical Disk / Physical Disk: % Free Space

But identifying disk bottleneck pertaining to time can be little tricky.  Following are the list of few disk parts that can create bottleneck pertaining to time.

a.       I/O Bus

b.      Device Bus

c.       Disk Controller

d.      Head Stack Assembly

e.       Disk Adapter

Over here, I will not discuss about isolating bottleneck present in the part of disk but overall disk.  Once it is confirmed that bottleneck is present in disk then later it can be diagnosed further to find which part of the disk is a bottleneck.

For identifying bottleneck present in the disk, plot graph against following counters:

a.       Logical / Physical Disk: Average Disk Queue Length

b.      Logical / Physical Disk: Current Disk Queue Length

Disk bottlenecks appear as sustained rates of disk activity above 85% of a sample interval and as persistent disk queues greater than 2 per disk.

But it is highly likely that the disk bottleneck is because of scarcity of memory available.

To rule out this that disk bottleneck is not because of memory, plot graph against following additional counters:

            c. Memory: Page reads / sec.

Follows the pattern of “Current Disk Queue Length” against “Page reads / sec”.  If “Current Disk Queue Length” is resulting because of “Page reads / sec” then the bottleneck present is because of memory rather than disk.