A benchmark of PostgresSQL operations. There are many different test configurations, I picked two that are closest to what Phoronix used in their benchmark article. scaling factor of 100 with 1000 clients. The read operation followed by read/write. Looks like there are mostly a small number of runnable processes through up to 800 appear at one point.
Topdown overview shows a front-end bound process. The timing looks slightly different in this case than previous one and stretching longer. Perhap additional runs to get stability.
AMD metrics show average on-cpu of only 2.5. Some floating point code and many branches. Frontend latency appears to dominate and this would be a good benchmark to dig deeper on latency.
elapsed 863.433
on_cpu 0.157 # 2.52 / 16 cores
utime 459.890
stime 1713.587
nvcsw 12451489 # 8.45%
nivcsw 134887770 # 91.55%
inblock 8 # 0.01/sec
onblock 16504 # 19.11/sec
cpu-clock 11390349332903 # 11390.349 seconds
task-clock 11403070781759 # 11403.071 seconds
page faults 91224025 # 7999.953/sec
context switches 397659971 # 34873.060/sec
cpu migrations 38233072 # 3352.875/sec
major page faults 799 # 0.070/sec
minor page faults 91223226 # 7999.882/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 5355467484433 # 205.853 branches per 1000 inst
branch misses 292739782623 # 5.47% branch miss
conditional 3407698242743 # 130.985 conditional branches per 1000 inst
indirect 117052096702 # 4.499 indirect branches per 1000 inst
cpu-cycles 45652716363402 # 3.31 GHz
instructions 27213399658895 # 0.60 IPC
slots 90040636490742 #
retiring 9410107725001 # 10.5% (11.7%)
-- ucode 44883771195 # 0.0%
-- fastpath 9365223953806 # 10.4%
frontend 45468404154306 # 50.5% (56.6%)
-- latency 37635521319684 # 41.8%
-- bandwidth 7832882834622 # 8.7%
backend 24909401939477 # 27.7% (31.0%)
-- cpu 2220898318464 # 2.5%
-- memory 22688503621013 # 25.2%
speculation 605948968091 # 0.7% ( 0.8%)
-- branch mispredict 602767658468 # 0.7%
-- pipeline restart 3181309623 # 0.0%
smt-contention 9643199770382 # 10.7% ( 0.0%)
cpu-cycles 45879696338268 # 3.32 GHz
instructions 26657314107701 # 0.58 IPC
instructions 8759583463590 # 79.491 l2 access per 1000 inst
l2 hit from l1 611600487555 # 22.86% l2 miss
l2 miss from l1 107286230319 #
l2 hit from l2 pf 32790526002 #
l3 hit from l2 pf 23400169289 #
l3 miss from l2 pf 28515861814 #
instructions 8759544138410 # 80.986 float per 1000 inst
float 512 70 # 0.000 AVX-512 per 1000 inst
float 256 586 # 0.000 AVX-256 per 1000 inst
float 128 709400728133 # 80.986 AVX-128 per 1000 inst
float MMX 0 # 0.000 MMX per 1000 inst
float scalar 0 # 0.000 scalar per 1000 inst
Intel metrics
elapsed 2291.272
on_cpu 0.139 # 2.22 / 16 cores
utime 1836.851
stime 3255.948
nvcsw 7011767 # 1.36%
nivcsw 507821164 # 98.64%
inblock 7648 # 3.34/sec
onblock 5576 # 2.43/sec
cpu-clock 27441549967843 # 27441.550 seconds
task-clock 27449294125546 # 27449.294 seconds
page faults 237069157 # 8636.621/sec
context switches 1078148357 # 39277.817/sec
cpu migrations 23548431 # 857.888/sec
major page faults 2234 # 0.081/sec
minor page faults 237066923 # 8636.540/sec
alignment faults 0 # 0.000/sec
emulation faults 0 # 0.000/sec
branches 9692774844607 # 191.453 branches per 1000 inst
branch misses 57955491662 # 0.60% branch miss
conditional 9692775358495 # 191.453 conditional branches per 1000 inst
indirect 2345563682863 # 46.330 indirect branches per 1000 inst
slots 124996051720466 #
retiring 32187185782664 # 25.8% (25.8%)
-- ucode 4181245792684 # 3.3%
-- fastpath 28005939989980 # 22.4%
frontend 62982740884535 # 50.4% (50.4%)
-- latency 43903284077682 # 35.1%
-- bandwidth 19079456806853 # 15.3%
backend 26797147178785 # 21.4% (21.4%)
-- cpu 4765353392012 # 3.8%
-- memory 22031793786773 # 17.6%
speculation 3176735760047 # 2.5% ( 2.5%)
-- branch mispredict 2355089638929 # 1.9%
-- pipeline restart 821646121118 # 0.7%
smt-contention 0 # 0.0% ( 0.0%)
cpu-cycles 59299293721852 # 1.62 GHz
instructions 45253336959405 # 0.76 IPC
l2 access 2944393364322 # 93.519 l2 access per 1000 inst
l2 miss 859430772105 # 29.19% l2 miss
Process overview shows the many processes launched as clients to run pgbench along with the postgres benchmark. Here would also be useful to separate out topdown metrics for either.
6595 processes
120 pgbench 7466.61 25775.90
6086 postgres 6313.98 2805.83
68 clinfo 18.19 5.99
38 vulkaninfo 1.32 0.96
6 glxinfo:gdrv0 0.15 0.08
4 vulkani:disk$0 0.14 0.10
6 php 0.07 0.17
2 llvmpipe-0 0.07 0.05
2 llvmpipe-1 0.07 0.05
2 llvmpipe-10 0.07 0.05
2 llvmpipe-11 0.07 0.05
2 llvmpipe-12 0.07 0.05
2 llvmpipe-13 0.07 0.05
2 llvmpipe-14 0.07 0.05
2 llvmpipe-15 0.07 0.05
2 llvmpipe-2 0.07 0.05
2 llvmpipe-3 0.07 0.05
2 llvmpipe-4 0.07 0.05
2 llvmpipe-5 0.07 0.05
2 llvmpipe-6 0.07 0.05
2 llvmpipe-7 0.07 0.05
2 llvmpipe-8 0.07 0.05
2 llvmpipe-9 0.07 0.05
2 glxinfo 0.07 0.04
2 glxinfo:cs0 0.07 0.04
2 glxinfo:disk$0 0.07 0.04
2 glxinfo:sh0 0.07 0.04
2 glxinfo:shlo0 0.07 0.04
6 clang 0.04 0.08
3 rocminfo 0.03 0.00
1 lspci 0.00 0.03
1 ps 0.00 0.01
90 sh 0.00 0.00
13 gcc 0.00 0.00
12 pg_ctl 0.00 0.00
10 gsettings 0.00 0.00
8 stat 0.00 0.00
8 systemd-detect- 0.00 0.00
6 bc 0.00 0.00
6 createdb 0.00 0.00
6 llvm-link 0.00 0.00
6 sleep 0.00 0.00
5 phoronix-test-s 0.00 0.00
4 dropdb 0.00 0.00
4 gmain 0.00 0.00
2 cc 0.00 0.00
2 dconf worker 0.00 0.00
2 lscpu 0.00 0.00
2 uname 0.00 0.00
2 which 0.00 0.00
2 xset 0.00 0.00
1 date 0.00 0.00
1 dirname 0.00 0.00
1 dmesg 0.00 0.00
1 dmidecode 0.00 0.00
1 grep 0.00 0.00
1 ifconfig 0.00 0.00
1 ip 0.00 0.00
1 lsmod 0.00 0.00
1 mktemp 0.00 0.00
1 qdbus 0.00 0.00
1 readlink 0.00 0.00
1 realpath 0.00 0.00
1 sed 0.00 0.00
1 sort 0.00 0.00
1 stty 0.00 0.00
1 systemctl 0.00 0.00
1 template.sh 0.00 0.00
1 wc 0.00 0.00
1 xrandr 0.00 0.00
5 processes running
1037 maximum processes