-
Notifications
You must be signed in to change notification settings - Fork 264
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Performance / Energy regression in Alpine 3.19.0 #385
Comments
Any idea what is causing this? Exactly what components are tested in the alpine image? the C library? openssl? |
The reason for the energy regression is uknown. We have so far only pinned it down to that version number of the alpine docker image. Under https://github.com/green-coding-solutions/alpine-energy-regression/blob/main/README.md you can see what we are executing. It is this command: No syscalls should be done and if you strace the command it holds at a certain point and all further actions happen in userspace, which is the expected behaviour: ...
futex(0x56231280b2f0, FUTEX_WAKE_PRIVATE, 1) = 1
alarm(0) = 30
futex(0x762127443910, FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, 5657, NULL, FUTEX_BITSET_MATCH_ANY
# [No output for around 30 seconds]
futex(0x762126b4f910, FUTEX_WAIT_BITSET|FUTEX_CLOCK_REALTIME, 5659, NULL, FUTEX_BITSET_MATCH_ANY) = 0
alarm(0) = 0
write(1, "CPU speed:\n", 11CPU speed:
... Important note: the To drill deeper a next step would be a flamegraph, but I thought I open the ticket first to maybe get some insights from you who know the container better and can maybe point in a possible rewarding direction :) |
Also to add to this: I tried an # Alpine view
/ # ps
PID USER TIME COMMAND
1 root 0:00 /sbin/docker-init -- sysbench cpu run --cpu-max-prime=25000 --threads=8 --time=800 --events=0 --rate=0 --debug=off
7 root 2:52 sysbench cpu run --cpu-max-prime=25000 --threads=8 --time=800 --events=0 --rate=0 --debug=off
16 root 0:00 ash
23 root 0:00 ps
/ # lsof -p 7
1 /sbin/docker-init 0 /dev/null
1 /sbin/docker-init 1 pipe:[65798]
1 /sbin/docker-init 2 pipe:[65799]
7 /usr/bin/sysbench 0 /dev/null
7 /usr/bin/sysbench 1 pipe:[65798]
7 /usr/bin/sysbench 2 pipe:[65799]
16 /bin/busybox 0 /dev/pts/0
16 /bin/busybox 1 /dev/pts/0
16 /bin/busybox 2 /dev/pts/0
16 /bin/busybox 10 /dev/tty So somehow it does not show me the libraries opened. Under Linux the tool behaves differently and I can see that ssl and libc are loaded. So I guess it will also include ssl libraries under # Ubuntu view
$ sudo lsof -p 6091
COMMAND PID USER FD TYPE DEVICE SIZE/OFF NODE NAME
sysbench 6091 arne cwd DIR 259,5 4096 20447234 /home/arne
sysbench 6091 arne rtd DIR 259,5 4096 2 /
sysbench 6091 arne txt REG 259,5 233056 3969323 /usr/bin/sysbench
sysbench 6091 arne mem REG 259,5 526896 3960085 /usr/lib/x86_64-linux-gnu/libgmp.so.10.4.1
sysbench 6091 arne mem REG 259,5 289800 3960233 /usr/lib/x86_64-linux-gnu/libhogweed.so.6.4
sysbench 6091 arne mem REG 259,5 281000 3960459 /usr/lib/x86_64-linux-gnu/libnettle.so.8.4
sysbench 6091 arne mem REG 259,5 1743016 3960876 /usr/lib/x86_64-linux-gnu/libunistring.so.2.2.0
sysbench 6091 arne mem REG 259,5 2000320 3932373 /usr/lib/x86_64-linux-gnu/libgnutls.so.30.31.0
sysbench 6091 arne mem REG 259,5 2260296 3960798 /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.30
sysbench 6091 arne mem REG 259,5 47688 3959961 /usr/lib/x86_64-linux-gnu/libffi.so.8.1.0
sysbench 6091 arne mem REG 259,5 92312 3960813 /usr/lib/x86_64-linux-gnu/libtasn1.so.6.6.2
sysbench 6091 arne mem REG 259,5 129096 3960275 /usr/lib/x86_64-linux-gnu/libidn2.so.0.3.7
sysbench 6091 arne mem REG 259,5 1285888 3960532 /usr/lib/x86_64-linux-gnu/libp11-kit.so.0.3.0
sysbench 6091 arne mem REG 259,5 105392 3960724 /usr/lib/x86_64-linux-gnu/libsasl2.so.2.0.25
sysbench 6091 arne mem REG 259,5 63992 3936920 /usr/lib/x86_64-linux-gnu/liblber-2.5.so.0.1.12
sysbench 6091 arne mem REG 259,5 4455728 3934197 /usr/lib/x86_64-linux-gnu/libcrypto.so.3
sysbench 6091 arne mem REG 259,5 22600 3960332 /usr/lib/x86_64-linux-gnu/libkeyutils.so.1.9
sysbench 6091 arne mem REG 259,5 182864 3960330 /usr/lib/x86_64-linux-gnu/libk5crypto.so.3.1
sysbench 6091 arne mem REG 259,5 827936 3960338 /usr/lib/x86_64-linux-gnu/libkrb5.so.3.3
sysbench 6091 arne mem REG 259,5 376512 3936921 /usr/lib/x86_64-linux-gnu/libldap-2.5.so.0.1.12
sysbench 6091 arne mem REG 259,5 338648 3960150 /usr/lib/x86_64-linux-gnu/libgssapi_krb5.so.2.2
sysbench 6091 arne mem REG 259,5 2220400 3934299 /usr/lib/x86_64-linux-gnu/libc.so.6
sysbench 6091 arne mem REG 259,5 125488 3960024 /usr/lib/x86_64-linux-gnu/libgcc_s.so.1
sysbench 6091 arne mem REG 259,5 841808 3961075 /usr/lib/x86_64-linux-gnu/libzstd.so.1.4.8
sysbench 6091 arne mem REG 259,5 940560 3934304 /usr/lib/x86_64-linux-gnu/libm.so.6
sysbench 6091 arne mem REG 259,5 6834168 3969319 /usr/lib/x86_64-linux-gnu/libmysqlclient.so.21.2.36
sysbench 6091 arne mem REG 259,5 52016 3960340 /usr/lib/x86_64-linux-gnu/libkrb5support.so.0.1
sysbench 6091 arne mem REG 259,5 18504 3959801 /usr/lib/x86_64-linux-gnu/libcom_err.so.2.1
sysbench 6091 arne mem REG 259,5 14432 3934302 /usr/lib/x86_64-linux-gnu/libdl.so.2
sysbench 6091 arne mem REG 259,5 68552 3934347 /usr/lib/x86_64-linux-gnu/libresolv.so.2
sysbench 6091 arne mem REG 259,5 667864 3934198 /usr/lib/x86_64-linux-gnu/libssl.so.3
sysbench 6091 arne mem REG 259,5 14456 3969315 /usr/lib/x86_64-linux-gnu/libaio.so.1.0.1
sysbench 6091 arne mem REG 259,5 539424 3969317 /usr/lib/x86_64-linux-gnu/libluajit-5.1.so.2.1.0
sysbench 6091 arne mem REG 259,5 329912 3969321 /usr/lib/x86_64-linux-gnu/libpq.so.5.14
sysbench 6091 arne mem REG 259,5 240936 3934293 /usr/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2
sysbench 6091 arne 0u CHR 136,1 0t0 4 /dev/pts/1
sysbench 6091 arne 1u CHR 136,1 0t0 4 /dev/pts/1
sysbench 6091 arne 2u CHR 136,1 0t0 4 /dev/pts/1 |
Hey alpine linux team,
we are working in sustainable software engineering and specializing in open source energy measurement tooling.
We use the alpine docker image for a workload energy validation container we are using.
While doing measurements with it we noticed that for a reference workload the energy consumption went up and also the performance of the workload went down.
The change happened during the transition from
alpine:3.18.5
toalpine:3.19.0
We have documented all the details in this repository: https://github.com/green-coding-solutions/alpine-energy-regression/blob/main/README.md
It shows the difference of -1% Performance and +5% energy increase.
We have higlighted this with charts that come from one of our open source tools and also with
perf stat
commands to replicate.Super happy for feedback on this and we are also happy to further drill down if you have a component as a suspect that might be the culprit for the energy increase as the Alpine Update contains quite some libraries that we are not too familiar with
@ribalba - Looping you in
The text was updated successfully, but these errors were encountered: