WINDOWS 1002000 PROFESSIONAL RESOURCE KIT [Electronic resources] نسخه متنی

Investigating Disk Performance Problems

Several conditions must exist in order for you to determine that a disk bottleneck exists. These are a sustained rate of disk activity well above your baseline, persistent disk queues longer than two per disk, and the absence of a significant amount of paging. Without this combination of factors, it is unlikely that you have a bottleneck. However, if you suspect a disk-specific performance problem, monitor the following types of counters:

Paging counters (under the Memory object): Pages/sec, Page Reads/sec, Page Writes/sec

Usage counters: % Disk Time, % Disk Read Time, % Disk Write Time, % Idle Time, Disk Reads/sec, Disk Writes/sec, Disk Transfers/sec

Queue-length counters: Avg. Disk Queue Length, Avg. Disk Read Queue Length, Avg. Disk Write Queue Length, Current Disk Queue Length

Throughput counters: Disk Bytes/sec, Disk Read Bytes/sec, Disk Write Bytes/sec

NOTE

---

Although not reflected in disk activity, the rate of interrupt generation by your disk hardware can have a systemwide performance impact. Disk I/O can sometimes generate a sufficient number of interrupts to slow the performance of the processor. Although this does not constitute a "disk" bottleneck, it is a processor bottleneck caused by the disk system that can slow the responsiveness of the whole computer. For more information about monitoring disk interrupts and reducing their impact on system performance, see "Analyzing Processor Activity" in this book.

The following sections describe how you interpret the values of these counters to reveal or rule out a bottleneck.

Monitoring Paging

The symptoms of a memory shortage are similar to those of a disk bottleneck. When physical memory is scarce, the system starts writing the contents of memory to disk and reading in smaller blocks more frequently (this process is called paging). The less memory you have, the more the disk is used, resulting in a greater load on the disk system. Therefore, it's important to monitor memory counters along with disk counters when you suspect a performance problem with your disk system.

Monitor paging activity along with disk reading and writing, using the following counters:

Avg. Disk Queue Length

Disk Reads/sec

Disk Writes/sec

MemoryPages/sec

MemoryPage Reads/sec

MemoryPage Writes/sec

Figure 30.6 shows how a memory shortage can cause disk counters to indicate a problem.

Figure 30.6 Paging Activity Compared with Disk Activity

Notice that this figure shows a long disk queue, accompanied by a high rate of paging. Compare the number of page reads against the number of disk reads to see how many times the system accessed the disk to retrieve pages that were not found in memory, or to write pages to free up memory for new data coming in from the disk. When these values are high, the system does not have sufficient memory. Without inclusion of the memory counters to reveal this behavior, you might have assumed that the disk was inadequate. Upgrading the disk in this situation would not have cured the problem.

For more information about measuring memory and identifying memory shortages, see "Evaluating Memory and Cache Usage" in this book.

Monitoring Usage

A high-performance disk is capable of about 50 to 70 random or up to 160 sequential I/O operations per second. The components you are using, as well as the request size, bus speed, and other factors, determine your system's capacity. Judge the maximum acceptable usage that your system can sustain based on your experience. Disk-time values must not consistently exceed the rate you've established as your baseline for performance. Consistent values in the 70 percent to 85 percent range are a definite cause for concern. However, if a queue is developing, lower percentages might indicate a disk that is unable to handle the load. If you see extremely high rates of disk usage, investigate the factors that might be responsible. Monitoring Disk Transfers/sec (a counter with values equal to the sum of Disk Reads/sec and Disk Writes/sec) or the individual counters Disk Reads/sec and Disk Writes/sec can show you the number of requests for service by the disk; the values of these counters provide a measure of disk demand.

If your workloads consist of random bursts of high activity, you might see high activity rates followed by long periods of idle time. If you only look at the average counter values with these types of workload, it can appear that your disk isn't busy even though it was bottlenecked during those bursts of high activity. To determine how well your disk system is handling these bursts, sample at short intervals when the activity occurs.

NOTE

---

The disk time counters can yield inaccurate values when multiple disks are in use. You can compensate for this by monitoring % Idle Time and comparing its value with the values reported by the % Disk Time, % Disk Read Time, and % Disk Write Time counters.

Figure 30.7 depicts maximum disk usage and the development of a queue.

Figure 30.7 High Disk Time Values

Monitoring Queue Length

To determine the number of I/O requests queued for service, track Avg. Disk Queue Length for LogicalDisk or PhysicalDisk. Notice that this might overstate the true length of the queue, because the counter includes both queued and in-service requests. If the value of Avg. Disk Queue Length exceeds twice the number of spindles, then you are likely developing a bottleneck. With a volume set, a queue that is never shorter than the number of active physical disks indicates that you are developing a bottleneck.

Figure 30.8 shows a disk bottleneck with high disk usage and a long queue.

Figure 30.8 High Disk Usage and a Long Queue