How to share bandwidth equally between processes

bhe69 · 2025-10-06 21:23:10

Hello everyone,

ever since I switched to linux I noticed that a single process downloading a file will take up all available bandwidth (which is wanted), but leaves none for other processes. I cant remember I ever experienced something like this on windows.

For example while I download a file with pacman or wget, browsing with firefox is nearly impossible, as hardly any data is transferred from the servers.
The default setting of pacman (downloading 5 packages in parallel) sometimes even chokes itself, as pacman aborts, being not able to download further packages:

 linux-lts-6.12.48-1-x86_64                                                                138.2 MiB   963 KiB/s 02:27 [########################################################################] 100%
 Total ( 8/71)                                                                             329.0 MiB  2.24 MiB/s 02:27 [########################################################################] 100%
error: failed retrieving file 'webkitgtk-6.0-2.50.0-1-x86_64.pkg.tar.zst.sig' from geo.mirror.pkgbuild.com : Operation too slow. Less than 1 bytes/sec transferred the last 10 seconds
error: failed retrieving file 'webkit2gtk-4.1-2.50.0-1-x86_64.pkg.tar.zst.sig' from geo.mirror.pkgbuild.com : Operation too slow. Less than 1 bytes/sec transferred the last 10 seconds
warning: failed to retrieve some files
error: failed to commit transaction (failed to retrieve some files)
Errors occurred, no packages were upgraded.

Is there a setting to share the bandwidth between processes more equally? I'm aware I can limit the bandwidth of pacman/wget to a fixed value, but I'd rather want to use all bandwidth, but still give other processes the chance to download some data as well.

Thanks for your help!

seth · 2025-10-07 15:08:56

That's not normal behavior nor does it make a lot of sense.
Do you use some firewall that might perform process based traffic shaping?
Sure you're not just running into a slow mirror?
Check your system journal - the more likely scenario is that the network collapses under pressure (you might see disconnects, firmware crashes etc et pp)

bhe69 · 2025-10-08 20:43:01

No, I'm not using a firewall and the mirrors are pretty fast. The transfer speed of 2.24 MiB/s is approximately the bandwidth my internet provider promises.

While I downloaded the kernel package today I checked the system log, The only message I could find during this time was:

perf: interrupt took too long (2532 > 2500), lowering kernel.perf_event_max_sample_rate to 78900

seth · 2025-10-08 21:45:37

Please trigger such event (connection breakdown) and then post your complete system journal for the boot:

sudo journalctl -b | curl -F 'file=@-' 0x0.st

bhe69 · 2025-10-20 21:45:40

So far I wasn't able to trigger the error again (although it happened several times in the past). I will reply once I get the error again.

bhe69 · 2025-11-10 20:20:33

Ok, today it happened again. You can find journalctl here: https://0x0.st/K9kd.txt

Console output is:

$ sudo pacman -Syu
[sudo] password for user: 
:: Synchronizing package databases...
 core is up to date
 extra is up to date
 multilib is up to date
:: Starting full system upgrade...
resolving dependencies...
looking for conflicting packages...

warning: insufficient columns available for table display
Packages (34) boost-libs-1.89.0-3  enchant-2.8.12-2  gspell-1.14.1-2
              harfbuzz-12.2.0-2  harfbuzz-icu-12.2.0-2  icu-78.1-1
              iso-codes-4.19.0-1  lib32-harfbuzz-12.2.0-2  libcdr-0.1.8-3
              libe-book-0.1.3-19  libical-3.0.20-3  libmspub-0.1.4-18
              libnautilus-extension-49.1-3  libqxp-0.0.2-14
              libreoffice-still-25.2.7-2  libvisio-0.1.10-2  libxcrypt-4.5.1-1
              libxcrypt-compat-4.5.1-1  libxml2-2.15.1-3  libzmf-0.0.2-19
              luajit-2.1.1762617240-1  php-8.4.14-2  python-pikepdf-10.0.1-1
              python-pillow-12.0.0-2  qt5-base-5.15.18+kde+r109-2
              raptor-2.0.16-9  systemd-258.2-1  systemd-libs-258.2-1
              systemd-sysvcompat-258.2-1  tinysparql-3.10.1-2
              vte-common-0.82.1-2  vte3-0.82.1-2  webkit2gtk-4.1-2.50.1-2
              webkitgtk-6.0-2.50.1-2

Total Download Size:   269.42 MiB
Total Installed Size:  946.56 MiB
Net Upgrade Size:        1.35 MiB

:: Proceed with installation? [Y/n] 
:: Retrieving packages...
 icu-78.1-1-x86_64      11.7 MiB   520 KiB/s 00:23 [######################] 100%
 qt5-base-5.15.18...    13.1 MiB   441 KiB/s 00:31 [######################] 100%
 systemd-258.2-1-...     9.2 MiB   599 KiB/s 00:16 [######################] 100%
 webkit2gtk-4.1-2...    34.2 MiB   735 KiB/s 00:48 [######################] 100%
 webkitgtk-6.0-2....    34.2 MiB   521 KiB/s 01:07 [######################] 100%
 libreoffice-stil...   143.4 MiB   921 KiB/s 02:40 [######################] 100%
 Total ( 6/34)         269.4 MiB  1729 KiB/s 02:40 [######################] 100%
error: failed retrieving file 'qt5-base-5.15.18+kde+r109-2-x86_64.pkg.tar.zst.sig' from geo.mirror.pkgbuild.com : Operation too slow. Less than 1 bytes/sec transferred the last 10 seconds
warning: failed to retrieve some files
error: failed to commit transaction (failed to retrieve some files)
Errors occurred, no packages were upgraded.

Looking forward to hearing your ideas what went wrong :-)

seth · 2025-11-10 20:32:49

Journal is dud

Nov 10 20:28:37 box sudo[1592]:      user : TTY=pts/0 ; PWD=/home/user ; USER=root ; COMMAND=/usr/bin/pacman -Syu --ignore chromium
Nov 10 20:28:37 box sudo[1592]: pam_unix(sudo:session): session opened for user root(uid=0) by user(uid=1000)
Nov 10 20:28:50 box systemd-resolved[272]: Clock change detected. Flushing caches.
Nov 10 20:28:55 box systemd[1]: Starting Time & Date Service...
Nov 10 20:28:55 box systemd[1]: Started Time & Date Service.
…
Nov 10 20:29:25 box systemd[1]: systemd-timedated.service: Deactivated successfully.
Nov 10 20:29:57 box systemd-resolved[272]: Clock change detected. Flushing caches.
Nov 10 20:31:21 box sudo[1592]: pam_unix(sudo:session): session closed for user root
Nov 10 20:32:04 box sudo[2312]:      user : TTY=pts/0 ; PWD=/home/user ; USER=root ; COMMAND=/usr/bin/journalctl -b
Nov 10 20:32:04 box sudo[2312]: pam_unix(sudo:session): session opened for user root(uid=0) by user(uid=1000)

Time jumps backwards.

Nov 10 21:27:54 archlinux systemd[1]: Switching root.
Nov 10 21:27:54 archlinux systemd-journald[120]: Journal stopped
Nov 10 20:27:55 box systemd-journald[120]: Received SIGTERM from PID 1 (systemd).
Nov 10 20:27:55 box systemd[1]: RTC configured in localtime, applying delta of 60 minutes to system time.
Nov 10 20:27:55 box systemd[1]: systemd 258.1-1-arch running in system mode (+PAM +AUDIT -SELINUX +APPARMOR -IMA +IPE +SMACK +SECCOMP +GCRYPT +GNUTLS +OPENSSL +ACL +BLKID +CURL +ELFUTILS +FIDO2 +IDN2 -IDN +IPTC +KMOD +LIBCRYPTSETUP +LIBCRYPTSETUP_PLUGINS +LIBFDISK +PCRE2 +PWQUALITY +P11KIT +QRENCODE +TPM2 +BZIP2 +LZ4 +XZ +ZLIB +ZSTD +BPF_FRAMEWORK +BTF +XKBCOMMON +UTMP -SYSVINIT +LIBARCHIVE)
Nov 10 20:27:55 box systemd[1]: Detected architecture x86-64.

probably by A LOT - 1h, to be more precise.

Nov 10 20:27:56 box ntfs-3g[361]: Mounted /dev/sda8 (Read-Only, label "SONSTIGES", NTFS 3.1)
Nov 10 20:27:56 box ntfs-3g[360]: Mounted /dev/sda6 (Read-Only, label "SPIELE", NTFS 3.1)

So porn is now "sontiges"?

The cause of the time warp is https://wiki.archlinux.org/title/Dual_b … e_standard and I'd not be surprised if that's what's throwing the connection off (ssl/tls is time sensitive)

Also make sure to see the 3rd link below. Mandatory.
Disable it (it's NOT the BIOS setting!) and reboot windows and linux twice for voodo reasons.

bhe69 · 2025-11-15 22:48:35

I switched the BIOS time to UTC and will keep on observing if it continues.

Arch Linux

#1 2025-10-06 21:23:10

How to share bandwidth equally between processes

#2 2025-10-07 15:08:56

Re: How to share bandwidth equally between processes

#3 2025-10-08 20:43:01

Re: How to share bandwidth equally between processes

#4 2025-10-08 21:45:37

Re: How to share bandwidth equally between processes

#5 2025-10-20 21:45:40

Re: How to share bandwidth equally between processes

#6 2025-11-10 20:20:33

Re: How to share bandwidth equally between processes

#7 2025-11-10 20:32:49

Re: How to share bandwidth equally between processes

#8 2025-11-15 22:48:35

Re: How to share bandwidth equally between processes

Board footer