Temporal Workloads

The SCSI temporal workload supports both iSCSI and FC protocols in a single workload. The protocol is based on the test bed configuration. If both FC and iSCSI exist in the same test bed the default key performance indicator (IOPs, Throughput and Latency) charts are only for the first protocol in the test bed. The other metrics are in the all charts section.

The workload does not appear in the default library because it can only be created from the workload analysis performed on workload data imported into the Workload Data Importer.

Data Parameters

Block temporal workloads support all data parameters available in WorkloadWisdom, including the following implementation for the deduplication option of data reduction.

The following deduplication settings control how many duplicate data patterns are used:

  • Dedup Ratio. Generates data pattern over the course of the workload test run. Dedup Ratio is unique as it generates data pattern over the course of the test that yields a specified deduplication ratio.

  • Data needing to be written is. Specifies how much of the data pattern should be completely unique (that is, never repeated) versus the amount of data patterns that are duplicates (that is, repeatable data patterns) throughout the workload test run. File workloads do not have Data needing to be written is.

2019-11-15_09-09-33.png

The implementation of the deduplication option for temporal workloads is different from the non-temporal workloads, in that you have the option to have no data pattern that is completely unique by setting the Unique percentage to 0%. If it is set to 0, it means you want all data patterns to be repeating patterns (except for the first pass). The workload test generates a pool of 100 unique but repeatable data patterns, and reuses these patterns throughout the workload test run to yield the desired Dedup Ratio. When the initial pool of 100 unique duplicates is exhausted, another pool of 100 unique duplicates is generated.

For example, if the Dedup Ratio is set to 5:1, then the first pool of 100 unique duplicates is generated 5 times to yield a ratio of 5:1, and then a second pool of 100 unique duplicates is generated 5 times to maintain the ratio of 5:1, and so on. If the dedup ratio contains a decimal some of the patterns are written more than the integer portion and others equal to it to arrive at the ratio specified. Setting the Unique percentage to 50% with a 5:1 ratio causes each duplicate data pattern to be written 9 times along with one data pattern that is only used once to arrive at a 10:2 or 5:1 ratio.

Location

As with hotspot workloads, the SCSI temporal workload provides the ability to access some portions of the LUNs more frequently than others. It also enables those locations to grow or shrink over time as well as move.

Each concurrent worker is working on only one location at a time and only one LUN for the duration of the test. During the test the worker changes which region it is working on based on the setting to update the offset and size.

2019-11-15_09-13-07.png

Initial Offset determines where the region begins at the beginning of the test. The offset location for the region moves linearly for the duration of the test until reaching the Final Offset at the end of the test. Similarly, the size of the region is the Initial Size specified and grows or shrink linearly to until the Final Size is reached at the end of the test. Both settings can be either based on a percentage of the LUN or a size specified independent of the LUN size. The Access % determines how frequently the workload is accessing the region as it moves from region to region based on the frequency specified.

Many array architectures are designed to take advantage of hotspot activity that is common to many real-world workloads. Systems that do dynamic tier management rely on this behavior. Some systems might also consider hotspots when doing garbage collection or performing other maintenance tasks. Very few workloads are completely random across and entire LUN so it is important include some level of hotspots when testing.

To run the FC / iSCSI Temporal Workload with the Hot Spots functionality “turned off”, create one Hot Spot entry for Read and for Write, and for each entry, set Initial Offset and Final Offset to 0%, Initial Size and Final Size to 100%, and Access to 100%.

2019-11-15_09-13-57.png

While the Frequency setting does not materially make a difference in this case since there are no hot spots, you do have to set it to a value greater than 0.

2019-11-15_09-15-07.png

Load Properties

Load Properties for SCSI temporal workloads are somewhat different than for other workloads. The IOPs are automatically changing over time with pattern observed in the production workload. There is no need to set it.

The number of asynchronous requests was also moved to this section as it is closely related to the number of concurrent workers in determining how many requests can be outstanding at any point in time. This queue can be determined by the number of concurrent workers and how many total requests can be set up to 2048. Setting the combined value higher than 2048 artificially increases latency but might be needed to cover all LUNs.

2019-11-15_09-16-23.png

The workload can be scaled for “what if” scenarios and this value can be used in conjunction with the iteration.

When you enable MPIO in a workload (which implies the Test Bed has MPIO enabled), the Load Properties settings will treat the Workload Generator Ports that belong to the same MPIO Group as if they were a single logical Workload Generator Port.

2019-11-15_09-18-17.png

In the above example, each of the 4 LUNs will be accessed by 2 workers concurrently, and if the load is 400MB/s, then a total of 400MB/s will be generated from the MPIO Group. Exactly how much of the 400MB/s is generated by each of the two physical Workload Generator Ports that make up the same MPIO Group depends on the MPIO Policy that is set. If it is round-robin, then each Port will generate 200MB/s, all else being equal. If it is failover, then the primary port will generate 400MB/s.