Open jtaleric opened 3 months ago
For bare-metal Kepler testing I've been applying two loads ,one for cpu and another for memory.
For CPU I like to record the 'events per second' throughput result using this syntax: $ sysbench cpu run --time=120 --threads=$(nproc) | awk '/events per second:/{printf "%.2f\n", $NF}'
For MEMORY the key metric recorded is 'transferred' throughput using this syntax: $ sysbench memory run --time=120 --threads=$(nproc) --memory-scope=global --memory-total-size=1T --memory-access-mode=rnd | awk -F '(' '/transferred /{print substr($NF, 1, length($NF)-1)}'
Using this issue to track some feedback on the current Kepler performance workload improvements.