Visualizing the Production Workload Data

Summary View displays your analysis results and provides an overview of the production workload data. Summary View also provides an overview of the type of workload that would be created if you selected only a single summary workload. When you select a time window on IOPs, Latency, or Throughput chart in the Access Pattern section, the numbers on IOPs, Latency, Throughput, and Errors are changed on the corresponded values.

Read/Write Mix: Displays the read/write ratio observed in the workload data. This value is used to determine read/write ratio in created workloads.

Block Size Distribution: Displays the distribution of observed request sizes and their average IOPs by size. This information provides you insight into the block sizes that are configured in the servers and applications communicating with the storage infrastructure.

Command Mix: Displays the ratio of all observed commands. When troubleshooting, it is useful to determine if there are any unexpected commands or if the command ratio is higher than expected. Sometimes it is helpful just to see that a particular command is present, as some commands can be quite disruptive or are not seen in production environments.

Average Block Size: Gives a good indication of the changing relationship between IOPs and Throughput. Batch activities, transactions such as video streaming, or applications configured with large block sizes typically have higher average request sizes.

Latency vs. IOPs: Displays if latency is limiting IOPs or if IOPs are causing latency. The information is displayed as a scatter plot with a trend line. If the trend line ramps up to the right, the amount of IOPs is impacting the latency. The steeper the slope the greater the impact. Often there might be more than one apparent trend as different request sizes take longer. Intervals with fewer IO of large size tend to have higher latency. If the line is trending up this is generally an indication that the latency is caused by the IOPs, but is significantly limiting it. If the line is trending down or flat, this indicates that the array is having no trouble keeping up with the requested workload.

Latency vs. Request Size: Displays how much the request sizes are impacting performance. All things being equal, a block size that is twice as big as another should take twice as long. You might see that is not the case and it can be normal. For example, you might see that 4KB reads take the same amount of time as 8KB reads. This is often an indication that the array is configured with an 8KB block size, and is accessing 8KB, when only 4KB is requested. It might be a good idea to consider aligning the server block size settings with the arrays by changing one or the other.

Click Create Workload to generate a summary workload from production workload data.

The following options are provided as displayed below, if during the Workload Data Importer process you selected multiple workloads to create and if the analysis policy you chose supports the workload types. So it is not a defect if you do not see this pop-up window at all.

These options allow you to determine the amount of load to generate when running the modeled workload in a test environment. The option you select is reflected in the Load Properties section of the workload configuration on the next page. Regardless of which option you select, you can customize any of the parameters that are exposed in the workload configuration page before starting a test. See Creating a New Workload Test. The option you select modifies a set of defaults in the workload configuration page so that you do not have to manually enter the information.

If during the Workload Data Importer process you selected the Composite Workload option and if the analysis policy supports it, then the Workloads Components View tab is available. Workloads Components View displays the discrete workloads identified by WorkloadWisdom’s workload analytics. The key characteristics for each identified workload are displayed, including how many Target/LUNs or SDFs (Source Destination Filesystem ID). Quality % indicates the relative accuracy of workload identification determined by the analytics.

You can rename workloads inline based on their observed behavior, number of LUNs or shares, or on the suspected type of workload present, like writes to Redo LUNs of a database or shares.

Each component workload is automatically named with the first word from the production workload data name along with the IOPs, average request size, and number of LUNs or SDFs that the workload represents and should be run against. The workloads are automatically sorted by the descending value of IOPs.

From the Workloads Components View tab, click Create Workload to generate a composite workload from production workload data. A workload is created for each of the selected components. The workload is named according to what was entered in the box in the first column. A composite workload is also created that includes all of the selected workloads. Once the workloads have been created, you are redirected to the newly created composite workload. Each component workloads is mapped to a test bed connection with the number of LUNs or SDFs clustered from the production workload.

Depending on the source of the data that is imported not all of the information might be available on the Analysis Results page.

For example, the following screen shot displays a case where the storage array does not collect enough information from its production workload data to be analyzed more thoroughly.