Examining and Tuning Disk Performance

Data Collection

Monitor disk counters along with counters from other objects. The following is a list of recommended counters.

LogicalDisk\% Free Space
PhysicalDisk\Disk Reads/sec
PhysicalDisk\Disk Writes/sec
PhysicalDisk\Avg.Disk Queue Length
Memory\Available Bytes
Memory\Cache Bytes
Memory\Pages/sec
Processor(All_Instances)\% Processor Time
System\Processor Queue Length

For server computers, add networking counters. These include the following:

Network Segment\% Net Utilization (you must enable the Network Monitor Driver protocol to use this counter).
Transmission counters, such as Segments/sec, Datagrams/sec, or Frames/sec (for the object installed by the networking protocol in use).
Application counters such as Server\Bytes Transmitted/sec or Server\Bytes Received/sec for the Server service.

Figure 8.5 depicts a typical display for collecting overall system performance data.

Enlarge figure

Figure 8.5 Counter Configuration for Baseline Monitoring

Observe activity at various times of day over a range of intervals, starting with one day, one week, one month, and so on. Over time a pattern develops, and you can see that the data tends to fall consistently within a particular range of values — that resulting range is your baseline.

You can monitor for short intervals such as two to five seconds, if your workload is characterized by random bursts of heavy activity. Otherwise 60-second intervals are adequate. If system demands fluctuate during the day, you might want to take shorter samples during periods of heaviest activity and longer samples when activity is tapering off.

For best results during monitoring, try to isolate the disk so that workload unrelated to your test does not affect your results. If you are logging performance to a disk that you are monitoring, values for the disk reflect a small amount of writing activity for that logging.