Introduction

Using Iometer correctly can demonstrate the value of ioTurbine.

Iometer is a powerful open source tool for measuring I/O performance under controlled loads.  It generates workload, and measures its impact.  In addition, it's now an open source tool, so can be freely downloaded.

Iometer is based on a client-server model in which one interface manages multiple workers performing IO operations.  Iometer allows the user to specify test parameters, allowing flexibility in configuring tests appropriate for the application.  Those parameters include the type of access (sequential or random), the size of the transfer data and the percentage of reads. 

Prerequisites

Before installing and using Iometer on a Windows VM, make sure you have fulfilled the following requirements:

• One or more ioMemory Device(s) installed in the ESXi host.
• Compatible version of the ioMemory VSL software installed on the ESXi host.
• One or more virtual machine (with a guest OS compatible with ioTurbine) running on the ESXi host.
• ioTurbine installed and configured.
  • ioMemory device(s) configured to cache data VM data.

Iometer Installation

Iometer is an open source tool that can be freely downloaded from the Iometer project site (http://www.Iometer.org).

1. Select a windows download from the download page (http://www.Iometer.org/doc/downloads.html).
2. Once the download is completed, you can use the wizard to install Iometer.
   
3. Follow the steps to install Iometer. You will have to accept the license terms the first time you run Iometer.

Recommended Settings

All benchmarking tests are, by definition, contrived; artificial tests cannot perfectly represent real application performance. Some test parameters return sub-standard results because they are designed for legacy hardware configurations and under-use the software and/or hardware.

However, with the correct settings, you can maximize the benchmarking of the hardware and software. The following settings closely represent real-world application of ioTurbine.

Parameter	Value
Number of CPUs	1
Number of Workers	3
Test Time	3 hours
Transfer Rate	64K

Test Setup

There are several steps to setting up an Iometer test. Details about how to configure each item in Iometer can be found in the Iometer documentation.

1. Go to the first tab (Disk Targets).
   
   Expand the name of the host (in the above example it's FUSIONT4425B) to see the workers.
2. Add workers (3).
3. Select the disk target for each worker.
4. Go to the Access Specifications tab to set up the test.
5. Set the test specification to the parameters you want. See the section on recommended settings for suggestions. Refer to the Iometer document if you want to modify any of the default tests. Be sure to set access specifications for all your workers.
   
   The following access specifications should be set to use the recommended settings.
   1. Select a 100% read specifier to modify.
   2. Change the random/sequential distribution to 100% random.
   3. Change the transfer request size to 64K
6. If you are planning a timed test, go to the Test Setup tab, and set your time.
   
   
7. When you're satisfied with your set up, press the green flag to start the test.

Results

The result display tab shows the results of your test. For final results, click the start of test radio button.

There are several statistics shown. The most useful are IOPS (I/O operations per second), MB per second, and average response time. You can demonstrate the value of caching with IOturbine by running two tests with the recommended settings, the first without caching and the second one with it.

You can calculate the value of caching as percent as follows:

(Number with caching – number without caching) / number without caching. For example, for IOPS without caching of 100 and 175 with caching you'd get

(175 - 100)/100 = 75/100, or a 75% improvement.

The recommended settings should provide a solid demonstration of the benefits reaped from using ioTurbine software.

In addition to the direct results in Iometer, you can also see the benefits in the performance graphs of the management server.

This graph shows the percent of cache hits. The hits go to 100% fairly quickly.

For Iometer, testing the Read and Write tabs (percent and IOPS) are irrelevant because this setup tested Iometer with 100% reads.