Export (0) Print
Expand All

Using System Monitor and Exchange Load Generator Data

 

Topic Last Modified: 2007-10-18

Exchange Load Generator performance counters represent only a partial view of the operation of a system. To fully understand how the system works, analyze the information in the Windows and Exchange counters by running Windows® System Monitor during the Exchange Load Generator test.

Use System Monitor data to do the following:

  • Track how closely various resources are to saturation, and adjust subsequent Exchange Load Generator tests accordingly.

  • Use the server performance data to identify bottlenecks and resources that are in high demand, and consider re-tuning the system and then running the same test again.

You should log the System Monitor data from both the Exchange server and the Exchange Load Generator client on a single, separate computer. For more information about consolidating your log files on a single computer, see How to Consolidate System Monitor Data.

Consider the following questions when looking at Exchange Load Generator test results and the performance data from System Monitor. To additionally help you answer these questions, view the log files in Exchange Load Generator and System Monitor.

  • Were there any errors in the test? Did they significantly affect the load?

    You can use both the LoadGen.log file and Windows Event Viewer to locate any test errors. To determine the validity of the test, examine the type and quantity of errors.

  • Did the client computers create the expected load?

    If too many users are simulated per client computer, Exchange Load Generator generates less than the expected load. To determine whether too many users are causing Exchange Load Generator to generate a lighter load, you can designate a control client that has one-fourth or one-eighth the number of users as the other client computers as a control for your test. If the per-user rate of events that Exchange Load Generator generates is different for the control client and the other clients, your experiment is invalid because the workload that is generated is incorrect. If your other client computers are bottlenecked, they will generate less than the expected requests. Additionally, if you compare the response times between the regular clients and the control client, you can help determine whether the number of users per client computer is affecting the results.

  • What was the actual load on the server with regard to transactions per user per day? Does this match the expected value?

    Make sure that you answer these two questions. To better understand the accuracy of the test and the performance characteristics of a server, consider the actual work that the server performed. If the server performed a different amount of work than expected, your test represents a different workload than intended.

  • Were there any server resources that were saturated? Were there any software queues growing unbounded?

    The following table lists the system monitor counters that can help you determine how close to saturation your server's resources are. The expected values are only recommendations and may change depending on the goals of your test and the role of the server.

    noteNote:
    To use the System Monitor disk counters listed in the table, on Windows 2000 Server or Windows NT, you must run the command diskperf –y, and then restart the computer before simulation starts on each client and server participating in the test. By default, on Windows Server 2003, the disk counters are enabled so that you do not have to run the diskperf command.
noteNote:
The expected results for Exchange 2007 counters are in line with the latest Microsoft Operations Manager (MOM) rules for Exchange 2007.

 

Object and Counter Description Expected Result Applies to Version

Processor: % Processor Time (_Total)

Displays the total processor time.

As the load increases, this counter increases to almost 100 percent, at which point the processor is bottlenecked, and no additional rate increases are possible.

Both

MSExchangeIS: RPC Requests

Displays the number of incoming RPC requests from clients.

This value should be less than 50 at all times.

Both

MSExchangeIS: RPC Averaged Latency

Displays the average latency of the RPC requests.

This value should be less than 70 ms at all times.

Both

MSExchangeIS: VM Largest Block Size

Displays the largest virtual memory block size available in the Exchange store.

This value should be larger than 500 MB at all times.

Exchange 2003

MSExchangeIS: VM Total 16MB Free Blocks

Displays the number of 16-MB free blocks available in the Exchange store.

This value should be larger than 1 at all times.

Exchange 2003

MSExchangeIS: VM Total Free Blocks

Displays the number of free blocks available in the Exchange store.

This value should be larger than 1 at all times.

Exchange 2003

MSExchangeIS: VM Total Free Blocks Bytes

Displays the total size of all free blocks available in the Exchange store.

This value should be larger than 500 MB at all times.

Exchange 2003

Memory: Available Mbytes

Displays the available memory in MB on the server.

This value should be larger than 500 MB at all times.

Exchange 2003

Memory: Free System Page Table Entries

Displays the total number of PTEs being used in the system. If this number reaches zero, the system will be in an unstable state.

This value should never reach zero.

Both

Logical disk:

Average Disk Queue Length

Displays the average disk queue length.

The number for all drives should reach a steady state at some point with any client errors, for example.

Both

Logical disk:

Current Disk Queue Length

Displays the current disk queue.

The number for all drives should reach zero every minute or less.

Both

Logical disk:

Average Sec/Read and Average Sec/Write

Displays the disk latency.

The latency for all drives should be less than 50 ms at all times.

Both

MSExchangeIS Mailbox: Message Queued for Submission

Displays the current number of submitted messages that are not yet processed by transport.

Average queue size should not increase during the test. It should be less than 250 throughout the test.

Exchange 2007

MSExchangeIS Mailbox: Receive Queue Size

Displays the number of items waiting to move from the SMTP queue to the mailbox store.

The average number should be less than one percent of the number of users in the simulation. Average queue size should not increase during the test. It should remain relatively constant.

Both

MSExchangeTransport Queues( Total)\Active Mailbox Delivery Queue Length

Displays the number of items in the active mailbox queues.

The average queue size should not increase during the test. It should be less than 250 throughout the test.

Exchange 2007

MSExchangeTransport Queues( Total)\Active Remote Delivery Queue Length

Displays the number of items in the active remote delivery queues.

The average queue size should not increase during the test. It should be less than 250 throughout the test.

Exchange 2007

MSExchangeTransport DSN\Failure DSNs Total

Displays the number of failure delivery status notifications (DSNs) that have been generated.

This value should be zero throughout the test.

Exchange 2007

MSExchangeTransport Queues( Total)\Submission Queue Length

Displays the items in the submission queue.

Average queue size should not increase during the test. It should be less than 250 throughout the test.

Exchange 2007

Exchange Load Generator Engine: Task Queue Length

Displays the queue of pending tasks on an Exchange Load Generator client.

Average queue size should not increase during the test. It should be less than 100 throughout the test.

Both

MSExchangeIS mailbox: Messages Submitted/sec

Indicates the rate that messages are submitted by clients.

System Monitor data should match the Exchange Load Generator predicted value for message submission rate. The expected average rate is:

Messages submitted per minute = (0.1771) * (number of users)

Note   This rate varies during a test run because of users who are logging on and off and users replying to and forwarding messages.

Exchange 2007

MSExchangeIS mailbox: message recipients delivered per minute

Indicates the message delivery rate.

System Monitor data should match the Exchange Load Generator predicted value for message received rate.

Message recipients submitted per minute = (0.4375) * (number of users)

Exchange 2003

MSExchangeIS Mailbox: Send Queue Size

Displays the number of items awaiting mail delivery; that is, the number of items waiting to move from the mailbox store to the SMTP queue in Internet Information Services (IIS) for categorization. SMTP determines the appropriate server to which to send mail.

The average number should be less than one percent of the number of users in the simulation. Average queue size should not increase during the test. It should remain relatively constant.

Exchange 2003

SMTP server: Categorizer queue length

Displays the queue of items waiting to be categorized in IIS.

Average queue size should not increase during the test. It should remain relatively constant. It should be less than 10.

Exchange 2003

SMTP server: local queue length

Displays the queue of items waiting in IIS to be delivered to the local mailbox store after categorization.

Average queue size should not increase during the test. It should remain relatively constant. It should be less than 1,000 throughout the test.

Exchange 2003

SMTP server: remote queue length

Displays the queue of items in IIS waiting to be delivered on another server after categorization.

Average queue size should not increase during the test. It should be less than 1,000 throughout the test.

Exchange 2003

SMTP server: NDRs Generated

Displays the number of Non-Delivery Receipts sent.

This value should be zero throughout the test.

Exchange 2003

MSExchangeIS mailbox: messages submitted per minute

Indicates the message submission rate.

System Monitor data should match the Exchange Load Generator predicted value for message submission rate. The expected average rate is:

Messages submitted per minute = (0.1771) * (number of users)

Note   This rate varies during a test run because of users who are logging on and off and users replying to and forwarding messages.

Exchange 2003

MSExchangeIS mailbox: message recipients delivered per minute

Indicates the message delivery rate.

System Monitor data should match the Exchange Load Generator predicted value for message received rate.

Message recipients submitted per minute = (0.4375) * (number of users)

Exchange 2003

MSExchangeIS Transport Driver: TempTable Current

This is the queue of items waiting to go from the Exchange store to IIS.

Average queue size should not increase during the test. It should be less than 1,000 throughout the test.

Exchange 2003

noteNote:
The System Monitor counter Average Delivery Time in the MSExchangeIS Mailbox performance object will be incorrect when you run Exchange Load Generator. This counter includes delivery times for messages that were delivered during Initialization. Delivery Time is the time, in milliseconds, between an Exchange server receiving the message from the client, and an Exchange server delivering the message to a mailbox.
For Initialization messages, the time that an Exchange server received the message from the client is the timestamp from the original archived message. Therefore, the delivery time for Initialization messages will frequently be more than one year. Because Initialization message delivery times are included in the Average Delivery Time counter, the average will be incorrect.
 
Was this page helpful?
(1500 characters remaining)
Thank you for your feedback

Community Additions

ADD
Show:
© 2014 Microsoft