The real value of the Processor Utilization option lies in the fact that NetWare is not a pre-emptive Network Operating System (NOS). Without pre-emption NOS, CPU control remains with executing processes until such time as the process performs a system request which will force control to be passed back to the task scheduler; or the process intentionally relinquishes control through a thread switch or sleep function.

This problem is compounded by the need to design NLMs using a different methodology from that of single-tasking DOS workstation programs. Unfortunately, more NLM developers originate from the single tasking DOS conditions than from the multi-tasking OS/2 and UNIX environments. Even those NLM developers who are familiar with other multi-tasking environments are typically used to the pre-emptive nature of those environments which will forcibly insure CPU time slicing.

These tendencies can cause some NLM developers to focus too much on the task at hand; rather than on the overall environment. The result is what is often referred to as 'ill-behaved' NLMs. The information provided by the Processor Utilization option can be used to identify the source of potentially excessive resource loads on the Server during periods of peak processor activity.

Upon selecting the Processor Utilization option from the Monitor main menu, a sub-menu will appear listing the active processes at that particular moment. Here it should be noted that the list of processes is only updated upon selection of the Processor Utilization option from the main menu. Therefore, threads which are initiated after the Processor Utilization option is selected, threads that are subsequently activated by previously loaded NLMs or as a result of additional NLMs being loaded, will not appear in the list. Moreover, those threads which were active at the time the option was selected but have since terminated remain in the list despite being non-existent.

The processes shown in the list represent both active threads and interrupts associated with the Server's present state. Some of the threads are specific to NetWare, while others are associated with various Novell and third party NLMs which are loaded at that time. While the sub-menu of active processes is displayed, it is possible to look at the activity associated with a single process by positioning the highlighted bar over the desired process and pressing enter. Multiple processes can be selected by pressing the F5 function key whenever the highlighted bar is positioned over a desired process to mark it; and then pressing the enter key once all desired processes have been marked. All of the listed processes can be selected by pressing the F3 function key.

Upon selecting one or more processes to view, a screen will appear with each of the processes listed on a separate line, along with three columns of data with the headings Time, Count, and Load. Following the list of processes are the histogram values: The maximum possible time slice; The overhead associated with tracking the selected information; And, the adjusted or weighted value which will be used as the sample for calculating Processor Utilization. The data associated with each process and the histogram overhead are updated once per second.

The first column (Time) indicates the amount of CPU time spent executing code within the process (i.e., the thread or Interrupt Service Routine). The higher this value, the more CPU time that the process actually consumed. This value is directly related to the load imposed on the Server by the process.

The second column (Count) shows how many times the process was executed over the last second. This value, if taken alone, can be confusing. The higher this value, the more often the process was executed which is representative of a level of activity. In the same respect, each count also represents a point where the process has, through design or by force, relinquished control so that other tasks can be run. Therefore, if other tasks had needed CPU time, they would have had the opportunity to allocate time during each such relinquish of control by the process.

The third column (Load) is a calculation of the relative CPU load imposed by the associated process. The Load percentage is based upon the adjusted or weighted maximum processing time.

Some processes poll for tasks by checking a flag each time executed. If the flag is set, the process will perform the task and then resume polling for tasks. Otherwise, the process will relinquish control to other processes via a ThreadSwitch() or ThreadSwitchWithDelay() function. Once control returns to this process, it will then resume polling for tasks.

These processes may appear to impose heavy loads upon the Server although the utilization reported by the Monitor NLM remains low. In reality, the process is really doing nothing more than spinning while looking for something to do. As more processes become active, the Load associated with this process will be reduced since its continuous relinquishing of control allows the other processes to allocate CPU time. In this instance, the values can be confusing since the process will actually reduce its Load the more active the Server becomes.

An example of such a process can be found in NetWare's internal Polling Process (NetWare v3, it has been more appropriately renamed to the Idle Loop in NetWare v4). NetWare's Polling Process (or Idle Loop) continuously checks for tasks to do. Upon discovering a task needing service, it spawns the appropriate process and resumes its polling for tasks. Thus, the Polling Process (or Idle Loop) ideally should be the most active process on a NetWare Server: The one constantly looking for something to do, versus those actually performing tasks.

Key things to watch for are processes which consume a higher than average amount of time and infrequently relinquish control (i.e., high Time and low Count). Such processes tend to indicate an NLM which is not necessarily well behaved.