JPS5971555A - Performance monitoring system - Google Patents

Abstract

PURPOSE:To pick up a bottle neck with a low overhead, by measuring selectively only an event relating to a job in response to on/off of a monitor mask at the switching of a job executed by an operating system. CONSTITUTION:Hooks 6, 8, 10 for asynchronous events and a hook 12 for synchronous events are incorporated in processing routines 5, 7, 9, 11 of the operating system 1. The action of each hook is controlled by turning on/off an asynchronous event monitor mask 13 and a synchronous event monitor mask 14. Although the mask 13 remains on state while a monitor is in operation, the mode is switched respectively into the total event measuring mode and the asynchronous event measuring mode by turning on/off the mask 14. The switching is performed by a mask switching routine 17 in a monitor program 2. When the control is handed over from the hooks 10, 12 to a data collecting routine 18, the information in the control table in the system 1 is collected in response to the event and set it to a buffer 19.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a monitoring method for dynamic performance characteristics in a computer system, and particularly to a monitoring method suitable for identifying performance bottlenecks.

[Prior art]

In order to fully explain the conventional monitoring system, the basic mechanism for monitoring will first be explained with reference to FIG. Monitoring all processing of a computer system, that is, events that occur during job processing? The following mechanism is used to record sequentially. An instruction called a hook 3 for transferring control to the monitor program 2 is embedded in the program of the operating system 1 to be monitored. When control reaches the hook position during operating system processing, if the corresponding bit in the control register provided as hardware (called monitor mask 4) is on, an interrupt will occur. Then control is transferred to monitor program 2. If monitor mask 4 is off, no interrupt is generated and the next instruction is all executed. That is, the monitor mask 4 in the control register serves as a switch for data collection in the monitor program 2.

Conventionally, this mechanism is used without any connection to the operating system control scheme. As shown in the flowchart of FIG. 2, the mask is first turned on to begin monitoring, and when finished, the mask is turned off. Therefore, in the case of a multi-programming computer system, interruptions for monitoring the processing of all jobs being executed occur.

Therefore, when the number of jobs being executed is large, monitoring overhead increases, resulting in a decrease in system performance. As a result, the disadvantages of conventional monitoring methods are that it becomes impossible to monitor the system while it is running, and that the system itself changes slightly for monitoring purposes, making it impossible to grasp the exact situation. Ta.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide a monitoring method that eliminates paper overhead, which is a performance bottleneck during operation of a multi-programming computer system.

[Summary of the Invention] There are various methods for achieving the above object, and one method is to selectively monitor by limiting the jobs to be monitored. In order to achieve this with low overhead, we thought it would be possible to turn all monitor masks on and off in synchronization with the operating system processing. That is, by turning on and off the monitor mask that controls all monitor interrupts when the operating system switches jobs to be executed, it is possible to reduce the number of monitor interrupts that occur and perform low-overhead monitoring.

〔Example〕

An embodiment of the present invention will be described below with reference to FIG.

This embodiment is a monitor that collects data showing the breakdown of the response time of a time-sharing system and identifies performance bottlenecks. Similarly, the terminal in this embodiment generally corresponds to a job, and this embodiment is similarly applicable to batch processing and the like other than time sharing systems.

The monitor of this embodiment monitors a limited number of terminals from among the terminals undergoing time sharing processing. Each individual terminal has one name within the operating system.
corresponds to one address space. Address space is a memory space of virtual addresses created for each terminal. In order to monitor all processes related to the terminal to be measured, it is sufficient to collect all data only during processes related to the corresponding address space.

Next, a method for selectively measuring only events related to the measurement target address space will be explained, but first, the two measurement modes that this monitor has will be explained. This monitor has an asynchronous event measurement mode and an all-event measurement mode. The asynchronous event measurement mode is a mode during execution of an address space that is not a measurement target, and only measures events such as input/output interrupts that occur asynchronously with the execution of the address space. The all-event measurement mode is a mode during execution of the measurement target address space, and in addition to asynchronous events, it also measures synchronous events that occur during execution of the address space, such as a supervisor call that calls an operating system routine. This monitor switches between these two modes and performs selective measurements.

FIG. 3 shows this mode switching and data collection method. Operating system 1 processing routine 5
,7,9. Hooks for asynchronous events during if 6.8
．． 10 and a hook 12 is embedded for synchronization events. The activity of each hook is controlled by turning on and off the monitor mask 13 for asynchronous events and the monitor mask 14 for synchronous events. Since asynchronous events are always measured, the asynchronous event monitor mask 13 remains on while the monitor is running. Then, by turning on and off the synchronous event monitor mask 14, the mode is switched to the all-event measurement mode and the asynchronous event measurement mode, respectively. This switching is performed by a mask switching routine 17 in the monitor program 2. The CPU dispatcher 7 within the operating system 1 is a central processing unit (C
It has the function of allocating PUJ to address space, but when control is passed to this CPU dispatcher 7, hook 8
Control then moves to a mask switching routine 17.

Then, by checking the sub-terminal table 16, the CPU determines that if the identifier of the destination address space is in the table 16, it is the address space to be measured, and turns on the synchronous event monitor mask 14. to enter all-event measurement mode. When Table 16 is reached, turn it off and enter the asynchronous event 11 measurement mode. Measurement terminal job offer 16 has a set of measurement terminal name and its corresponding address space identifier. The terminal name is set to the name of the specified measurement terminal when the monitor is started. The address space identifier is set as follows when the user logs on (operation to start using the terminal) or logs off (operation to end using the terminal) at the terminal.

At logon/logoff, control is passed from the logon/logoff processing routine 5 to the measurement end recruitment update routine 15.
If the name of the terminal that logged on or logged off is in the measurement terminal recruitment 16, it is a measurement terminal, so the address space identifier of that terminal is set in the table 16; otherwise, it is not set.

Now you can make correspondence between measurement terminal and address space.

When control is passed to the data collection routine 18 from a hook 10.12 in a routine 9.11 within the operating system, the operating system 1
Collect all the information in the control table in 1.
The data are collected into one record and set in the buffer 19. Each time buffer 9 becomes full, monitor program 2 outputs the entire contents of buffer 9 to file 20.

From the data of this monitor, you can calculate the details of response time, such as command response time, input/output execution waiting time, and memory reservation time. Although it is made of gold for a long time, it becomes a candidate for a bolt neck in the breakdown item system performance.

〔Effect of the invention〕

The effect of the present invention is manifested in the reduction of monitoring overhead, and when the number of terminals operating at the same time is n, the overhead of the monitor of this embodiment is l/n compared to the conventional monitor.

According to the present invention, in a multi-programming computer system, performance bottlenecks can be extracted with low overhead. Assuming that the multiple production money of a job is n, the overhead is about 1/n compared to the conventional method. It also has the effect of making it possible to perform more detailed monitoring of a specific job than was previously possible due to overhead.

[Brief explanation of drawings]

Figure 1 is a basic structure diagram of monitoring, Figure 2 is a flowchart of a conventional monitoring method, and Figure 3 is an embodiment of the present invention.
4) on the book chart. DESCRIPTION OF SYMBOLS 1... Operating system, 2... Monitor program, 3... Hook, 4... Monitor mask, Logon/logon processing routine, 6... Logon/logoff hook, 7... ...CPU dispatcher, 8.Dispatcher hook, 9.Asynchronous processing routine, 10.Asynchronous event hook, 11.
...Synchronization processing routine, 12...Synchronization event hook, 1
3...Monitor mask for asynchronous events, 14...Monitor mask for synchronous events, 15...Discount end job update routine, 16...Total (j!lI terminal table, job offer... Mask switching routine, 1B...data collection routine,
19...Buffer, 20...File. Agent: Patent Attorney Toshiyuki Usuda ) times

Claims (1)

[Claims] 1. In performance monitoring to obtain a detailed breakdown of processing time during actual operation of a multi-programming computer system, several specifications can be set by turning on and off the monitor mask when switching jobs executed by the operating system. A performance monitoring method that is characterized by selectively measuring only events related to jobs that have been performed.