Q: 20 ms latency is hardcoded?


A: There are two sets of latency thresholds defined in the Jetstress xml configuration file: One is strict for a test short than or equal to 6 hours. The other is lenient for a test longer than 6 hours (it is also called a stress test).

 

Excerped from Jetstress chm help file — Enables you to specify the period for performance sample gathering. If you specify less than or equal to six hours, the resulting performance log will contain six hours of performance samples. The performance test becomes a stress test when the duration is longer than six hours so that it uses a lenient set of the disk I/O latency thresholds.

 

 


The latency thresholds for Exchange 2003 still apply for Exchange 2007.


 

Editing these thresholds is undocumented. I guess that it is ONLY available for some corner cases.

 

– <Thresholds Name=”Strict”>
      <AverageDatabaseReadLatency>20</AverageDatabaseReadLatency>
      <MaximumDatabaseReadLatency>50</MaximumDatabaseReadLatency>
      <AverageLogWriteLatency>10</AverageLogWriteLatency>
      <MaximumLogWriteLatency>50</MaximumLogWriteLatency>
  </Thresholds>
– <Thresholds Name=”Lenient” TestRunTypes=”Stress”>
      <AverageDatabaseReadLatency>20</AverageDatabaseReadLatency>
      <MaximumDatabaseReadLatency>100</MaximumDatabaseReadLatency>
      <AverageLogWriteLatency>10</AverageLogWriteLatency>
      <MaximumLogWriteLatency>100</MaximumLogWriteLatency>
  </Thresholds>

 

CAUTION: They are editable through the xml configuration file.

You have to use an extra CAUTION in editing thresholds,

since using big thresholds can break down your hardware.

 

 

Hyungmin Lee [MSFT]

This posting is provided “AS IS” with no warranties, and confers no rights.

Comments (2)

  1. napobo3 says:

    When I tried to run Jetstress on one host, 3000 mailboxes, 2 IOPS, the tuning phase finished with

    "Tuning process couldn’t find suitable parameters, please use the suppress tuning to specify parameters."

    Does it mean that the test of 6000 mailboxes, 1 IOPS will give the same result ?

    Thank you,

    — leon

  2. MSDN Archive says:

    In tuning perspective, the following test cases are the same input. So, a comparable result will be the output.

    Test Case #1:

    3000 mailboxes, 2 IOPS, mailbox size 1024MB

    Test Case #2:

    6000 mailboxes, 1 IOPS, mailbox size 512MB

    If you use a database size smaller than 80% of the LUN size, there is a likelyhood of being compromised to a short-stroking effect.

    i.e. 6000 total IOPS at 20% of the LUN size doesn’t imply 6000 total IOPS at 80% of the LUN size.