A stress engine of various Linux operations with 39 workload tests around various system calls. As might be expected there is a somewhat differing set of workloads for different system calls and similarly at least one that tries having a large number of runnable processes and some that test interrupt behavior.
Topdown overview shows similar behavior all over the map. To dive deeper, makes sense to separate out the workloads.
Composite is shown in AMD topdown metrics. Overall keeping most cores busy. Some floating point and a reasonable number of branches.
elapsed 5084.757
on_cpu 0.773 # 12.37 / 16 cores
utime 32335.341
stime 30543.925
nvcsw 798829363 # 24.59%
nivcsw 2449890031 # 75.41%
inblock 6920 # 1.36/sec
onblock 22937248 # 4510.98/sec
cpu-clock 63338126482931 # 63338.126 seconds
task-clock 63370492886426 # 63370.493 seconds
page faults 252282205 # 3981.067/sec
context switches 3247234343 # 51242.056/sec
cpu migrations 40594272 # 640.586/sec
major page faults 3885 # 0.061/sec
minor page faults 244989996 # 3865.995/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 33287849051067 # 134.232 branches per 1000 inst
branch misses 2040381135392 # 6.13% branch miss
conditional 21738326981131 # 87.659 conditional branches per 1000 inst
indirect 1043706930671 # 4.209 indirect branches per 1000 inst
cpu-cycles 246020356645369 # 3.02 GHz
instructions 247522262682045 # 1.01 IPC
slots 494262022605330 #
retiring 91346010020862 # 18.5% (23.6%)
-- ucode 458614951576 # 0.1%
-- fastpath 90887395069286 # 18.4%
frontend 175331383055920 # 35.5% (45.4%)
-- latency 138264249623772 # 28.0%
-- bandwidth 37067133432148 # 7.5%
backend 116225434441614 # 23.5% (30.1%)
-- cpu 36027639621668 # 7.3%
-- memory 80197794819946 # 16.2%
speculation 3425655658401 # 0.7% ( 0.9%)
-- branch mispredict 3407508332244 # 0.7%
-- pipeline restart 18147326157 # 0.0%
smt-contention 107922040366956 # 21.8% ( 0.0%)
cpu-cycles 237754882893348 # 3.02 GHz
instructions 245483193483317 # 1.03 IPC
instructions 82870905019605 # 29.832 l2 access per 1000 inst
l2 hit from l1 1876625583798 # 11.86% l2 miss
l2 miss from l1 121920128803 #
l2 hit from l2 pf 424202227051 #
l3 hit from l2 pf 109437666801 #
l3 miss from l2 pf 61904909671 #
instructions 82821378194925 # 95.294 float per 1000 inst
float 512 180 # 0.000 AVX-512 per 1000 inst
float 256 3512658061 # 0.042 AVX-256 per 1000 inst
float 128 7888690609108 # 95.249 AVX-128 per 1000 inst
float MMX 0 # 0.000 MMX per 1000 inst
float scalar 141197153 # 0.002 scalar per 1000 inst
Intel metrics
elapsed 6757.530
on_cpu 0.785 # 12.56 / 16 cores
utime 53318.572
stime 31555.783
nvcsw 2058548634 # 26.09%
nivcsw 5832029530 # 73.91%
inblock 637688 # 94.37/sec
onblock 13248472 # 1960.55/sec
cpu-clock 85035232641948 # 85035.233 seconds
task-clock 85064973177120 # 85064.973 seconds
page faults 371711301 # 4369.734/sec
context switches 7887342199 # 92721.386/sec
cpu migrations 68032834 # 799.775/sec
major page faults 12513 # 0.147/sec
minor page faults 361797738 # 4253.193/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 46012903714132 # 129.178 branches per 1000 inst
branch misses 283085258308 # 0.62% branch miss
conditional 46013293141460 # 129.180 conditional branches per 1000 inst
indirect 14151704748033 # 39.730 indirect branches per 1000 inst
slots 517436504170508 #
retiring 227053908476335 # 43.9% (43.9%)
-- ucode 27953978000842 # 5.4%
-- fastpath 199099930475493 # 38.5%
frontend 133851851528919 # 25.9% (25.9%)
-- latency 66874983273868 # 12.9%
-- bandwidth 66976868255051 # 12.9%
backend 141357582855630 # 27.3% (27.3%)
-- cpu 29262841118363 # 5.7%
-- memory 112094741737267 # 21.7%
speculation 14979447431469 # 2.9% ( 2.9%)
-- branch mispredict 13354893130669 # 2.6%
-- pipeline restart 1624554300800 # 0.3%
smt-contention 0 # 0.0% ( 0.0%)
cpu-cycles 194962391661297 # 1.65 GHz
instructions 292457779985404 # 1.50 IPC
l2 access 4054293120327 # 17.627 l2 access per 1000 inst
l2 miss 729690506741 # 18.00% l2 miss
Process summary, dies after 23 tests, there are more than five million processes forked – so this is also a good stress test for the tools.