02_cache/01_latency

Hands-on: Measuring memory latency by LMbench

• Both of C/C++ and Fortran users will perform the common benchmarks, although LMbench is written by C.

How to execute

1. Edit a job script

• Before trying this hands-on, you need to do 00_lmbench to compile LMbench.
• A job script to execute the program is located in results/.
• Edit BINDIR variable in results/task.sh before the execution.
  • You need to write your installed location of LMbench there. In this exercises, we use lat_mem_rd.

2. Run program

• You can run the program either:

## To run as a batch job
$ cd results
$ pjsub task.sh 
## Or, to run in an interactive job
$ cd results
$ bash task.sh 

• The job in the Exercise will be completed within 5 minutes.
  • For safety, we set the elapsed time of the job script as 7 minutes.

Exercises A

• E1: Examine the hardware specification of cache in A64FX, focusing on the following points.
  • How many layers in cache are there?
  • What about the size of each cache?
  • Whether is a cache shared with cores or not?
• E2: Check the command-line option in lat_mem_rd.
  • You do not need to take care of the first option (-P 1).
  • The second option (256) means the maximum memory size, in unit of MiB.
    • This value should be larger than the size of the last-level cache at least.
  • The remaining numbers indicate strides in unit of bytes to access memory address.
• E3: Check the output of lat_mem_rd, depending on the values of stride. In particular, find the points of memory size in which the latency abruptly changes.
  • The first column indicates memory size (MiB),
  • The second one indicates the measured latency (ns).