LAT (Latency) Test

webinar

Summary:

Latency (LAT) measures average and maximum response times using RND (Random) IOs at specified Block Sizes and R/W (Read/Write) mixes taken under PTS Steady State condition.  LAT measurements are displayed in a summary Table and in 2D and 3D Plots. 

Test Setup:

The general LAT test sequence is: 

  1. PURGE the device.
  2. Perform the necessary Workload Independent Preconditioning (WIPC). 
  3. Perform Workload Dependent Preconditioning (WDPC), which is also the actual test stimulus, until Steady State Criteria as set forth in the Specification is reached.  The Steady State Criteria calls for maximum data excursion and maximum slope excursion to be within 20% and 10% of the average, respectively.  Both average and maximum latency is captured.

Note that the Steady State Measurement Window (the portion of data that is used to determine conformance to Steady State) is made up of 5 “Rounds”.   A "Round" in the context of the LAT test is a single sweep through the 9 element Block Size x R/W mix matrix (3 Block Sizes: 0.5, 4, 8 KiB and 3 R/W Mixes: 0/100, 65/35 and 100/0).  Each Block Size R/W combination is tested for one minute resulting in a single Round test time of 9 minutes.

Benefits of LATENCY Test:

  1. Provides Baseline:  The LAT test calls for the use of 1 total outstanding IO setting in the testing application driving the device.  The LAT results give the baseline round trip time for a single command from the time that the command is issued by the application to the time that command completion is acknowledged.  Assuming that the application and the associated software and driver stack do not become the bottleneck, the results here indicate the device’s performance without the benefit of a queue.

Average Versus Maximum Latency:  In addition to providing average latency (which in this specific case of total outstanding IO setting of 1 is just the inverse of the average IOPS), the LAT test also gives maximum latency values.  Comparing the maximum latency information between drives can be helpful to see if there might be processes within the drives that may cause maximum latency numbers to become large.    Note, however, that maximum latency behavior may be different when the drive is faced with more total outstanding IOs.