Job Management Partner 1/Performance Management - Agent Option for Platform Description, User's Guide and Reference

[Contents][Glossary][Index][Back][Next]


1.3.4 Operation on virtualized systems

The following explains how to use PFM - Agent for Platform to monitor performance on virtualized systems.

Organization of this subsection
(1) Objectives of operation on virtualized systems
(2) Monitoring resources on virtualized systems
(3) Usage examples of PFM - Agent for Platform on virtualized systems
(4) Precautions regarding usage of PFM - Agent for Platform on virtualized systems
(5) Virtualized system functionality and collected performance data
(6) Range of data collected for each virtualized system

(1) Objectives of operation on virtualized systems

The following explains the advantages of virtualized systems, and the objectives in running PFM - Agent for Platform on a virtualized system.

(a) Advantages of virtualized systems

Virtualized systems are attracting attention as a technology that makes the most of excess system resources. For example, with the current one-server, one-application paradigm, most servers consistently have excess resources. Specifically, except for times of server job congestion, usually only 5% to 7% of resources are used. Virtualized systems are a way to solve this issue.

Virtualized systems also have the following advantages:

(b) Objectives of performance monitoring

As discussed above, the advantages of a virtualized system are greatest when system resources are used in the proper amount. Virtualized systems cannot be used effectively if system resources are overconsumed, or on the other extreme, hardly used at all.

In situations like these, the status of resource consumption can be monitored to determine whether the current system is set up effectively.

At a basic level, PFM - Agent for Platform is designed to collect information about a single physical operating system. However, it can be used creatively to monitor performance information that depends on virtualized systems.

The main objectives of using PFM - Agent for Platform to monitor virtualized system performance monitoring are as follows:

In virtualized systems, specific bottlenecks might adversely affect performance for the entire virtualized system. The following might cause such bottlenecks:

The following explains methods for using PFM - Agent for Platform to monitor the performance of virtualized systems. Note that the thresholds shown here for performance monitoring are just reference values. Use the appropriate values based on the system configuration and usage.

(2) Monitoring resources on virtualized systems

(a) Overview

The following are the most important system resources to be monitored for performance:

In a virtualized system, some performance data for the above resources might not be collected or might become meaningless when changed dynamically. Performance changes for each OS, or by each virtualized system.

(b) Installing PFM - Agent for Platform on a virtualized system

On a virtualized system, each type of resource is basically managed by virtualized functionality (software or hardware), and configured by being allocated to different logical partitions. The managed resources include processors, memory, disks, and networks.

These resources can be obtained as performance data on the logical partition on which PFM - Agent for Platform is installed.

Some virtualized systems distinguish between guest operating systems and host operating systems, as is the case with VMWare, Virtage, and other software-based virtualizations. The host OS manages the guest OS, and the guest OS runs on the host OS. In such cases, PFM - Agent for Platform is installed on the guest OS to monitor performance.

PFM - Agent for Platform collects performance data on the installed logical partition, or guest OS.

Precautionary note:
Because the host OS for VMWare is a special OS for managing virtualized functionality, application operation is not guaranteed. Because Virtage creates an off-disk management area called an SVP frame, data collected by PFM - Agent for Platform cannot be saved as a database in this management area.

(c) Setting up PFM - Agent for Platform on a virtualized system

The procedures for collecting information for an installed instance of PFM - Agent for Platform are the same for virtualized systems. The existing settings can be used as is.

When a virtual machine is set up using a virtualized system, and PFM - Agent for Platform is installed on the machine, PFM - Agent for Platform collects performance data about the virtual machine.

(3) Usage examples of PFM - Agent for Platform on virtualized systems

The following gives simple examples of usage on a virtualized system.

(a) Determining whether to change virtualized system settings

If there are extra physical resources despite the consumption of a significant amount of resources on a specific guest OS or logical partition, we recommend that you change the virtualized system settings.

PFM - Agent for Platform can be used as a tool to determine these kinds of situations.

The following graph shows a case in which the allocation of processor resources is not sufficient.

Figure 1-5 Case in which processor resource allocation is not sufficient (during processor usage monitoring)

[Figure]

Because the processor usage for logical partition A is 100%, the allocation ratio is being switched. However, the processor usage for logical partition A does not decrease, while the processor for logical partition B is not fully used. In cases like this, change the allocation ratio settings.

The following graph shows a case in which the allocation of processor resources is sufficient.

Figure 1-6 Case in which processor resource allocation is sufficient (during processor usage monitoring)

[Figure]

The processor resource ratio is being switched while there are many processes in the queue for logical partition A. As a result, the number of processes in the queue for logical partition A is decreasing.

(b) Determining whether additional physical resources are needed

If physical resources always seem to be under load, consider adding another physical resource.

Note, however, that, even when a single guest OS or logical partition is under high load on a virtualized system, the system-wide load might seem low. Therefore, check whether all guest operating systems or all logical partitions are under high load.

When using PFM - Agent for Platform to check the load status in a virtualized system, you might require a different perspective from the one you use for an ordinary environment.

A virtualized system contains resources that can be changed dynamically, such as processors and memory. Therefore, when monitoring processor and memory usage, also monitor the items not easily affected by dynamic changes in resources, such as the current queue length and paging frequency.

The following graph gives an example of processor resources.

Figure 1-7 Case in which additional processor resources are recommended (during processor usage monitoring)

[Figure]

In the figure example, the allocation ratio is being switched due to the high load of processor usage for logical partition B, but usage is high for both logical partitions A and B. In this kind of case, physical resources are insufficient.

(4) Precautions regarding usage of PFM - Agent for Platform on virtualized systems

The performance data collected by PFM - Agent for Platform in a virtualized system is specific to each virtual machine. Therefore, PFM - Agent for Platform should be installed on each virtual machine.

Performance data about each virtual machine cannot be collected from the overall virtualized system, or from the management machine.

The following are precautions regarding the collection of information for each resource.

(a) Precautions regarding collection of processor information

The processor usage collected in a virtualized system might not yield the intended values. This is because the virtualized system divides up processor resources and dynamically allocates them to other logical partitions. For example, if significant amounts of processor resources are being used for a specific logical partition due to high load, only limited processor resources can sometimes be used on other logical partitions.

Figure 1-8 Case in which collected values differ due to dynamic resource changes

[Figure]

Note that if an application on a given logical partition is using a physical processor, jobs on other logical partitions are cause to wait due to operation. In this case, processor usage cannot be accurately ascertained.

(b) Precautions regarding collection of memory information

As with processors, information about memory usage on a virtualized system might also differ from the system-wide ratio, due to dynamic resource changes.

For example, when 256 MB of memory area is allocated, and only 64 MB of it is used, PFM - Agent for Platform displays a value of 25% for the memory usage allocation. In this case, when the memory allocation area is dynamically changed from 256 MB to 128 MB, PFM - Agent for Platform displays 50% for the memory usage. Keep in mind that an unexpected warnings might occur when a memory resource is specified for an alarm.

Figure 1-9 Case in which memory resources are built into an alarm

[Figure]

(c) Precautions regarding collection of disk information

As with processor resources, sometimes disk busy rates cannot be accurately ascertained. For example, on a virtualized system, some jobs might be using physical disk resources while others are waiting due to operation.

(d) Precautions regarding collection of network information

PFM - Agent for Platform measures only incoming and outgoing packets for network information. Therefore, operation on a virtualized system is the same as normal usage, but in some cases virtual network information is displayed for network information. This is not an issue, because PFM - Agent for Platform collects network information recognized by the OS running on the virtual machine.

(e) Precautions regarding collection of process information

Process-related performance data consists of items that show the process count and items that show the amount of resources used by processes. In both cases, since data on the logical partition is obtained for virtualized systems, this process information does not reflect the entire virtualized system.

PFM - Agent for Platform collects both the number of processes on each logical partition, and the amount of resources being used by the processes.

(5) Virtualized system functionality and collected performance data

In virtualized systems, the setup of provided resources differs depending on how the system is implemented. Also, resource changes might or might not require temporary stoppage of a logical partition. If a logical partition needs to be stopped, this should not affect much because PFM - Agent for Platform is also stopped. However, if a logical partition does not need to be stopped, and the system resources change dynamically, the performance data obtained by PFM - Agent for Platform might be significantly affected. In this case, we recommend collecting performance data not significantly affected by dynamic resource changes.

The following table lists the resources that can be collected for each virtualized system.

Table 1-34 Resources for each virtualized system

Virtualized system Resource
Processor count Memory Disk Network
VMware ESX Server 3.0 Required Required Required Required
Virtage Required Required -- Required

Legend:
Required: Requires temporary logical partition stoppage
--: Not supported

(a) VMware ESX Server 3.0

VMware is software that provides virtualized systems on the Intel architecture. Since it provides virtualized systems implemented in software, it is a general purpose implementation that can run flexibly on several environments.

l VMware ESX Server functionality

Like other virtualized systems, VMware provides the following functionality:

VMware manages combinations of the above functionality to create a virtual machine (VM). The created VM is recognized as a single host from other hosts, and can be treated the same as a physical machine.

Because multiple VMs can be created and run, operating systems like Linux and Windows can be run concurrently.

l VMware ESX Server configuration

VMware implements a virtualized system by creating a distinction between a host OS and a guest OS.

To set up a virtualized system:

  1. Install Windows, Linux, or another OS on the actual hardware.
  2. Install VMware on the installed host OS.
    Here, the OS on which VMware is installed is called the host OS.
  3. Use VMware to create a VM.
  4. Install other operating systems on the created VM.
    An operating system installed on the VM is called a guest OS.

As shown above, VMware uses a parent/child configuration to implement a virtualized system, in which the guest OS is executed on the host OS.

One feature of VMware ESX Server is that a specialized custom OS is used as the host OS instead of Windows or Linux. In other words, VMware ESX Server implements a virtualized system in which Windows or Linux is running on a specialized custom OS.

Depending on the type of VMware, the host OS can also be Windows or Linux instead of the custom OS.

Figure 1-10 VMware configuration

[Figure]

l Using PFM - Agent for Platform on VMware ESX Server

Keep the following in mind when using PFM - Agent for Platform on VMware ESX Server.

Installing PFM - Agent for Platform
To use PFM - Agent for Platform, install it on the guest OS. Because the host OS for VMware ESX Server is an optimized custom OS to implement virtualized systems with VMware, application operation is not guaranteed.

Processor information
In a virtualized system, if sufficient resources are not allocated to the host OS, impact will extend to the operation of the host OS. In particular, processor resources are likely to involve overload, and it might be difficult to determine whether the processor resources allocated to the VM are insufficient, or whether the host OS processor resources are insufficient.
VMware also has a feature that allows upper and lower limits to be set for processor usage allocation, which can cause wide swings in usage because resource switching is automatically performed within the set range.
If this seems to be the case, monitor items that are not easily impacted by dynamic changes to processor resources, such as the current queue length.

Memory
Be careful when setting an alarm for memory as a resource that can be changed dynamically. Even when the amount of memory used is fixed, if the allocated memory resources are reduced, memory usage will grow, and might cause alarms to be notified unintentionally.

Devices
Disk performance data is only for logical partition groups on the guest OS. System-wide performance data that includes the host OS cannot be collected. For the network adapter information, information about network adapters allocated to the guest OS is obtained.

(b) Virtage

l Virtage functionality

Like other virtualized systems, the functionality provided by Virtage includes the following:

Virtage manages the above functionality to implement a virtualized system. Whenever any resource is changed, the logical partition needs to be temporarily stopped.

l Virtage configuration

Like VMware, Virtage has a parent/child configuration.

To set up a virtualized system on Virtage:

  1. Install Virtage on the SVP.
  2. Start the installed instance of Virtage.
  3. Create and set up a virtual machine.
  4. Start the virtual machine.

l Using PFM - Agent for Platform on Virtage

Keep the following in mind when using PFM - Agent for Platform on Virtage.

Installing PFM - Agent for Platform
Because Virtage is configured in a special environment called an SVP frame, do not install PFM - Agent for Platform on an SVP frame.

Processor information
Because a virtualized system based on Virtage runs after the logical partition is set up, the number of processors is not changed while PFM - Agent for Platform is running. Accordingly, operation can be performed as usual, but the number of allocated processors might not match the number of processors that physically exist.

Memory
Because a virtualized system based on Virtage runs after the logical partition is set up, the amount of memory is not changed while PFM - Agent for Platform is running. Accordingly, operation can be performed as usual.

Devices
Disk performance data is only for logical partition groups. System-wide performance data that includes the host OS cannot be collected. For the network adapter information, information about network adapters allocated to the guest OS is obtained.

(6) Range of data collected for each virtualized system

The following table describes the range of data collected by PFM - Agent for Platform when it is used on a virtualized system.

Table 1-35 Range of data collected for each virtualized system

Record VMware and Virtage
PD Processes on the guest OS.
PD_APP Processes on the guest OS.
PD_DEV Devices for file system drivers and kernel drivers on the guest OS.
PD_ELOG Event log on the guest OS.
PD_GEND Used-defined performance data on the guest OS.
PD_PAGF Performance data about paging file instances on the guest OS.
PD_PDI Processes on the guest OS.
PD_PEND Processes on the guest OS.
PD_SVC Performance data about application services registered with the service control manager on the guest OS.
PD_UPD User-specified performance data on the guest OS.
PD_UPDB User-specified performance data on the guest OS.
PI System values on the guest OS.
PI_AD Active Directory performance data on the guest OS.
PI_BRSR --
PI_GENI User-defined performance data on the guest OS.
PI_ICM6 ICMP-related performance data on the guest OS.
PI_ICMP ICMP-related performance data on the guest OS.
PI_IP IP-related performance data on the guest OS.
PI_IP6 IP-related performance data on the guest OS.
PI_LOGD Performance data about the logical partitions allocated to the guest OS.
PI_NBT --
PI_NETI TCP/IP-related performance data on the guest OS.
PI_PCSR Information about processors allocated to the guest OS.
PI_PHYD Performance data about total values or average values for logical partitions allocated to the guest OS.
PI_SVRQ Performance data about the queue for the Server service provided by the guest OS.
PI_TCP TCP-related performance data on the guest OS.
PI_TCP6 TCP-related performance data on the guest OS.
PI_UDP UDP-related performance data on the guest OS.
PI_UDP6 UDP-related performance data on the guest OS.
PI_UPI User-specified performance data on the guest OS.
PI_UPIB User-specified performance data on the guest OS.
PI_WGRP Processes on the guest OS.
PI_WINS --

Legend:
--: N/A

[Contents][Back][Next]


[Trademarks]

All Rights Reserved. Copyright (C) 2009, Hitachi, Ltd.