Monitoring Node Performance
Applies To: Windows Compute Cluster Server 2003
Windows Compute Cluster Server 2003 provides two categories of performance objects:
Compute Cluster: cluster-specific performance counters
Compute Nodes: node specific performance counters
Compute Nodes and Compute Cluster performance objects are exposed on the head node only. Each performance object exposes a rich set of performance counters that can be used to monitor different characteristics of the cluster and nodes. All other operating system performance objects are exposed on all the nodes in the cluster.
The cluster administrator can monitor compute cluster and node performance from within the Compute Cluster Administrator by clicking Open System Monitor in Compute Cluster Administrator or by running a stand-alone copy of System Monitor (formerly called Performance Monitor) or any tool that can access System Monitor objects.
Counters can be added and customized views created using System Monitor within the Compute Cluster Administrator. Open System Monitor sessions and counters added to these sessions save saved on exit from the Compute Cluster Administrator.
To run System Monitor from the Compute Cluster Administrator
From the Node Management page of the Compute Cluster Administrator, right-click one or more nodes.
Select Open System Monitor.
The selected nodes are displayed beneath System Monitor on the navigation tree. The central details pane contains a standard System Monitor display. Basic information about each counter can be viewed by clicking the counter.
Eight predefined node counters are automatically displayed. These counters provide a quick look at node performance. These counters are:
Percentage processor time
Number of running jobs
Number of running tasks
Total number of processors
Number of processors in use
Number of processors idle
Number of serial tasks
Number of parallel tasks
You can add counters by right-clicking a counter in the lower pane of the central Details display and selecting Add Counters.
From the Add Counters dialog box, select the computer, the performance object, and the counters associated with that object that you want to add, and then click Add.
When finished, click Close. The selected counters are added to the counter display.
To stop monitoring a node, select the node beneath the System Monitor heading in the navigation Pane, and then click Close.
If you want to alter the way that information is displayed, right-click the node being monitored in the navigation pane and select Properties. Make the desired changes and click OK.
Note
If you change display options or add object counters, these changes are not saved between sessions.
If you are monitoring more than one node at a time, you can switch between sessions by selecting the node in the navigation pane.
Accessing compute cluster counters using System Monitor and other tools
You can view compute cluster counters by running Windows System Monitor as a stand-alone application or using any other tool that can access System Monitor objects. Windows System Monitor, called Performance Monitor in previous versions of Windows, is a tool you can use to monitor the resource utilization by system and application processes.
Note
Cluster counters apply to a single cluster. Node counters can apply to multiple nodes.
| Performance Object | Counter |
|---|---|
Compute Cluster |
Total Number of jobs in queue |
Compute Cluster |
Number of running jobs in queue |
Compute Cluster |
Number of pending jobs in queue |
Compute Cluster |
Number of not submitted jobs in queue |
Compute Cluster |
Number of failed jobs in queue |
Compute Cluster |
Number of canceled jobs in queue |
Compute Cluster |
Number of finished jobs in queue |
Compute Cluster |
Total number of tasks in queue |
Compute Cluster |
Number of running tasks in queue |
Compute Cluster |
Number of pending tasks in queue |
Compute Cluster |
Number of not submitted tasks in queue |
Compute Cluster |
Number of failed tasks in queue |
Compute Cluster |
Number of canceled tasks in queue |
Compute Cluster |
Number of finished tasks in queue |
Compute Cluster |
Total number of nodes |
Compute Cluster |
Number of idle nodes |
Compute Cluster |
Number of paused nodes |
Compute Cluster |
Number of unreachable nodes |
Compute Cluster |
Number of Pending for Approval nodes |
Compute Cluster |
Number of Ready nodes |
Compute Cluster |
Total number of processors in cluster |
Compute Cluster |
Number of processors in use in cluster |
Compute Cluster |
Number of processors idle in cluster |
Compute Cluster |
Number of processors unreachable in cluster |
Compute Nodes |
Node Manager State on node |
Compute Nodes |
Number of jobs running on node |
Compute Nodes |
Number of tasks running on node |
Compute Nodes |
Number of total processors on node |
Compute Nodes |
Number of processors in use on node |
Compute Nodes |
Number of processors idle on node |
Compute Nodes |
Number of tasks finished / hour |
Compute Nodes |
Number of tasks canceled / hour |
Compute Nodes |
Number of tasks failed / hour |
Compute Nodes |
Number of serial tasks running on node |
Compute Nodes |
Number of parallel tasks running on node |
For additional information about using Microsoft System Monitor, see System Monitor overview (https://go.microsoft.com/fwlink/?LinkId=63635)
If you are using Microsoft Operations Manager 2005, you can also use the Microsoft Windows Compute Cluster Server 2003 Management Pack to more easily manage your Windows compute cluster.
The Windows Compute Cluster Server 2003 Management Pack monitors the performance, health, and availability of Windows compute clusters. This management pack includes rules that:
Monitor critical Windows Compute Cluster Server 2003 services.
Capture operational errors.
Collect important Windows compute cluster performance counters.
When using this management pack, you are alerted to problems with Windows compute clusters as error conditions occur. By collecting and providing useful views of important cluster, node, job, and task performance, this management pack increases your ability to monitor compute cluster performance and to take action quickly when problems occur.
To download the Windows Compute Cluster Server 2003 Management Pack, see the Management Pack Catalog (https://go.microsoft.com/fwlink/?LinkId=77857).