|< Day Day Up >|| |
High processor values can mean either that your machine is handling the workload in a very efficient manner or that it is struggling to keep up. When a bottleneck occurs because a process’s threads need more processor cycles than are available, long processor queues build up, causing the system response to suffer. The two common causes of processor bottleneck are excess demand placed on the processor by CPU-bound programs and excess interrupts generated by drivers or subsystem components, such as disk or network components.
The minimum recommended counters for monitoring the server’s processor component for possible bottlenecks are the following.
System\ Processor Queue Length (all instances) A sustained queue length of more than 10 items per processor indicates a bottleneck.
Server Work Queues\ Queue Length A queue length of greater than four over a sustained period of time indicates possible processor congestion.
Processor\ Interrupts/Sec Use this counter to determine whether interrupt activity is causing a bottleneck. A dramatic increase in this counter value without a corresponding increase in system activity indicates a hardware problem. To resolve this problem, you need to find the network adapter or other device that is the source of the interrupts. Refer to the manufacturer’s specifications for the acceptable processor threshold; use 1000 interrupts per second as a starting point.
Processor\ % Interrupt Time Displays the percentage of time the processor spends receiving and servicing hardware interrupts during the sample interval. This value provides an indirect indication of the activities of devices that generate interrupts, such as disk drives, network adapters, and other peripheral devices. These devices interrupt the processor when they require attention or complete a task. Look for a dramatic increase in the value without a corresponding increase in system activity.
To monitor possible usage problems, use these counters:
Processor\ % Processor Time (all instances) Use this counter to discover a process that is using more than 85 percent of processor time. You might need to install an additional processor or upgrade to a faster one.
Processor\ % User Time Monitors the percentage of nonidle processor time that is spent in user mode. A high rate could indicate a need to upgrade or install additional processors. Use this counter in conjunction with Processor\ % Processor Time (all instances).
Processor\ % Privileged Time Monitors the percentage of nonidle processor time designated for hardware-manipulating drivers and operating system components. A high rate might be attributed to a large number of interrupts being generated by a device that is failing. Use this counter in conjunction with Processor\ % Processor Time (all instances).
Tips for Processor Monitoring
Try the following to resolve problems with the processor:
Upgrade the processor Upgrade to a faster processor, replace a failing one, or add another processor to the machine, especially if you are running multithreaded programs.
Adjust the workload of the system Distribute programs more efficiently among servers, or schedule programs to run at off-peak hours.
Manage processor affinity on multiprocessor computers Managing the processor affinity with respect to interrupts and process threads can improve performance because it reduces the number of processor cache flushes during thread movement from one processor to another.
You set affinity for a particular process or program when you assign it to a single processor to improve its performance, at the expense of the other processors. Be warned that when you dedicate a process or program to a processor, your other program threads might not be permitted to migrate to the least busy processor. You can set affinity in Task Manager, and it is available only on multiprocessor systems.
|< Day Day Up >|| |