Database Reference
In-Depth Information
Table 3-1.
(
continued
)
Object(lnstance[,lnstanceN])
Counter
Description
Value
Disk Transfers/sec
Rate of read/write operations
on disk
Maximum value dependent
on I/O subsystem
Disk Bytes/sec
Amount of data transfer to/
from per disk per second
Maximum value dependent
on I/O subsystem
Avg. Disk Sec/Read
Average time in ms to read
from disk
Average value < 10 ms, but
compare to baseline
Avg. Disk Sec/Write
Average time in ms to write
to disk
Average value < 10 ms, but
compare to baseline
The PhysicalDisk counters represent the activities on a physical disk. LogicalDisk counters represent logical
subunits (or partitions) created on a physical disk. If you create two partitions, say R: and S:, on a physical disk,
then you can monitor the disk activities of the individual logical disks using logical disk counters. However, because
a disk bottleneck ultimately occurs on the physical disk, not on the logical disk, it is usually preferable to use the
PhysicalDisk counters.
Note that for a hardware redundant array of independent disks (RAID) subsystem (see the “Using a RAID
Array” section for more on RAID), the counters treat the array as a single physical disk. For example, even if you
have ten disks in a RAID configuration, they will all be represented as one physical disk to the operating system, and
subsequently you will have only one set of PhysicalDisk counters for that RAID subsystem. The same point applies
to storage area network (SAN) disks (see the “Using a SAN System” section for specifics). Because of this, some of the
numbers represented in the previous table may be radically lower (or higher) than what your system can support.
Take all these numbers as general guidelines for monitoring your disks and adjust the numbers to account for the
fact that technology is constantly shifting, and you may see different performance as the hardware improves. We're
moving into more and more solid-state drives and even SSD arrays that make disk I/O operations orders of magnitude
faster. Where we're not moving in SSD, we're taking advantage of iSCSI interfaces. As you work with these types of
hardware, keep in mind that these numbers are more in line with platter-style disk drives and that those are fast
becoming obsolete.
% Disk Time
The % Disk Time counter monitors the percentage of time the disk is busy with read/write activities. This is a good
indicator of load but not a specific indicator of issues with performance. Record this information as part of the basic
baseline in order to compare values to understand when disk access is radically changing.
Current Disk Queue Length
Current Disk Queue Length is the number of requests outstanding on the disk subsystem at the time the performance
data is collected. It includes requests in service at the time of the snapshot. A disk subsystem will have only one disk
queue. With modern systems including RAID, SAN, and other types of arrays, there can be a large number of disks
and controllers facilitating the transfer of information to and from the disk. All this hardware makes measuring the
disk queue length less important than it was previously, but this measure is still useful as an indicator of load on the
system. You'll want to know when the queue length varies dramatically because it will be a possible sign of I/O issues.
But, unlike the old days, there is no way to provide a value that you can compare your system against. Instead, you
need to plan on capturing this information from your individual systems and using it as a comparison point over time.
