6. Appendix: Job profile listings¶

The following sections list job profiles that are made available to simplify benchmarking. Each profile contains a number of parameters (on the example of randread.fio):

rw=randread

Fio scenario. In this case, random reads.

size=8g

Dataset size. Needs to be at least twice the size of RAM.

numjobs=16

Number of parallel tasks. Needs to be equal to the number of CPU cores.

bs=4k

Data block size.

ioengine=libaio

Asynchronous library. Requires the parameter ‘direct=1’.

iodepth=32

Queue depth for asynchronous requests.

time_based

Instructs fio to loop the scenario for the duration of ‘runtime’ if it completes sooner or end the scenario if the time runs out.

runtime=60

Scenario duration (in seconds by default).

direct=1

This parameter forces data to be written directly to disk, without caching, and allow you to measure disk’s actual write performance.

filename_format=__testfile’$jobnum’

File name format. In this case it depends on the number of tasks.

thread

Task type. If not set, processes will be created instead of threads.

directory=/mnt/vstorage/benchmark_dir

Target directory for the dataset.