How to set CPU and IO priority for a systemd service? In partucular, how to set the highest values? (Assume that the service is idling most of the time and will absolutely NOT clog the rest of the system. Assume that the request is intentional and desired.)

I want a CPU-intensive process to have the lowest possible impact on other processes. I can use nice -19 for maximum niceness and chrt -i 0 to give the process a SCHED_IDLE priority. Which is more effective? Does it make any sense to combine the two like this? chrt -i 0 nice -19 my_command

OS is Debian 12. Got a brand new disk to use as a backup for a disk currently in use. For the initial rsync, my system starts to lag and freeze for maybe half a second at a time. This is especially noticeable when playing back audio. Load average jumps from ~5 to ~15 during the rsync. What is going on here? I know it's rsync because the instant I kill the process, audio playback goes back to normal. Trying to run with lowest cpu priority and lowest io priority like so: sudo nice -n 19 ionice -c 3 rsync -a --sparse --delete --info=progress2 /mnt/src/ /mnt/dest Running with sudo as a few files are owned by root and I want to be sure those get copied. How can I run rsync so my system doesn't freeze? Also I know it's not an issue with these specific disks, as I have tried other combinations of disks and I get the same issue. I also had this same problem on my older system, which used entirely different disks and motherboard.

I would like to start a process with a nice value of -20. This requires me to use a command like

sudo nice -n -20 matlab

However, this starts matlab as root too. Is there a way to have matlab as non-root? My current approach is

sudo nice -n -20 sudo -u myusername matlab

which to me looks like a hack. Is there a direct approach to do this?

I've noticed that all services and processes with my user in Ubuntu Server 22.04.4 LTS have a nice value of 19. I connect to the server through ssh, and my ssh sessions also have a nice value of 19. I didn't use the nice or renice commands. I'm searching for the configuration that produces this default nice value. So far, I've seen that /etc/security/limits.conf is the place to configure default nice value, but my current file is all commented out. Also I cannot find any reference to this nice value in the generic systemd user@.service or user.slice unit files. In desktop Ubuntu, the default nice value is 0, so I assume it must be configurable. Could you point me to the configuration file where this nice value of 19 is configured?

My nice is from GNU coreutils 9.1. I observed that nice -15 is equivalent to nice -n 15:

nice               # prints 0 for me, the base niceness is 0
     nice -n  15 nice   # prints 15, this is expected
sudo nice -n -15 nice   # prints -15, this is expected
     nice    -15 nice   # prints 15

-15 is a negative number. Why does it increment the niceness in the last example above? The manual (e.g. in Debian 12) does not explain this.

When I "flood" my CPU with 8 high priority (nice=-20) OS threads (the number of cores I have), operation becomes "halty" for obvious reasons, but is still usable. Note that when I say "high-priority thread" I mean a thread that was spawned by the same high-priority process. However doing something like 64 threads will completely make my PC unusable. What is the relationship between max priority threads and their distribution between cores? Can we figure out roughly how many threads I need to spawn to completely flood the CPU for a given nice value?

One of my server's users needs to launch CPU and I/O intensive jobs that are also quickly done (usually less than 30 seconds). They can be started by a user’s cron, a user’s shell or even Apache, PHP or MySQL processes forked/launched from root’s processes (MySQL and Apache run as root then forked processes set themselves as the corresponding user). Is there a way to nice an entire user and all its future processes owned by this user? **EDIT** Added more precision about the kind of processes targeted

I have a regular process that's not so important but will consume very much CPU power. I have another process which is really important, but it spends most of the time idle, but when it gets a job it really needs high computing power. I tried running with nice -20 ./low_priority_process and nice --20 ./high_priority_process but still the lower priority process consumes significant amount of CPU when the high priority process is in need. How can I run a process that will really yield or even auto-suspend when another process is using CPU power?

If I have server users who often forget to use the 'nice' command for their jobs on a web server, is it possible to change a setting so that the default priority for everything they launch is slightly higher than zero?

I want to set the process to the lowest possible scheduling on Linux. I came up with the following: nice -n 39 ionice -c 3 chrt -i 0 command Are there also other settings that one can set for a process to "lower" the process priority? Are these settings the "lowest" possible? More often than not, I do make -j$(nproc) or cmake or heavy tar processes that take full available I/O, memory and CPU on my machine. What is really frustrating, is that my mouse starts lagging. So I want to prevent it.

Debian 10. I am trying to render with blender in the background, like so: nice -n 19 blender -b --python my_script.py This renders to a file using the GPU. When rendering, I can see GPU usage spike to 100% with nvidia-smi, the system load goes up to around 3, and with top I can see cpu usage go to 110% for both blender and xorg. iostat shows almost zero iowait. At this point, my system becomes almost completely unresponsive. Windows take 10 seconds or so to get focus, etc. If I do ctrl+z, after 10 seconds or so the signal is processed and as soon as blender is suspended, my system instantly returns to normal. My end goal is to render in the background on my workstation while keeping my system responsive. How do I accomplish this?

I did this sudo nice -n -1 firefox to launch Firefox with higher priority and make all subprocesses that are executed at launch inherit that priority but it failed with an error message about Firefox can't be run as root in a regular user account. I need sudo to use negative nice values but the rest of the command should be run as my regular user. How can I use negative nice values but still run the program as my regular user? I don't want to use renice on already launched processes.

I run two extremely CPU-heavy (but not memory or I/O heavy) task on my Macbook Pro with 12 cores. This made the computer unusable for other tasks. I therefore reniced these processes to 20 and set Firefox to -1 (all other processes are 0 or lower), but Firefox is still extremely laggy. Shouldn't Firefox get all CPU time it requires when its nice-value is the lowest among all running processes?

I need to run some long and heavy commands, but at the same time, I'd like to keep my desktop system responsive. Examples: btrfs deduplication, btrfs balance, etc. I don't mind if such commands take longer to finish if I give them a lower priority, but my system should always be responsive. Using nice -n 19 and ionice -c 3 should solve my problem, but I'm not sure which command should come first for maximum benefit. * Option A:

nice -n 19 ionice -c 3 btrfs balance start --full-balance /

* Option B:

ionice -c 3 nice -n 19 btrfs balance start --full-balance /

Is there some subtle difference between options A and B? Are they equivalent perhaps?

- Long time ago (pre-2.6 kernels) on a very low end desktop, I had been happily following Con Kolivas suggestion to (quite heavily (-10)) renice the X server. (for better responsiveness) - With newer kernels, newer scheduler and smp systems, many had reported that even a 1 + or - variation in the default nice value was "*leading the rest of the system to suffer in quite obvious ways*". I went back to default. - Having no longer paid any attention to this, I recently got the surprise to discover /usr/bin/X (1.20.8) launched *-nolisten tcp -keeptty :0 -auth blahblah* running with a nice value of -1. (*1) ---------- **Q1 :** Is this (running the X server with a nice value of -1) nowadays considered the best option for a desktop mainly running SCHED_OTHER tasks under default nice value ?
Note *1 below would, in some very curious way indicate that, if not the best option, it is a deliberate will from freedesktop irrespective of any context (clients, hardware, kernel, cpu pinning…) **Q2 :** Considering that I am also running a window/compositing manager (kwin_x11) said achieving most of the job the X-server used to achieve, should I consider consequent to also renice it ? in a greater or lesser amount that the X-server ? (X consumes on average 1,5 cpu time kwin_x11 does) (*2) **Q3 :** Would it be pertinent to base the nice relative values of X / kwin_x11 on the relative values of their unvoluntary context switches ? (kwin_x11 gets (during some of my typical desktop workload) about 30 times more unvoluntary context switches than X) ---------- **Q4 :** More generally speaking, in a client/server SCHED_OTHER world, is there a rule of thumb regarding relative nice values between servers and clients apps ? And, considering this particular case (of X server/ X client), can I assume that the client will, most of the time, be waiting for the server's availability ?
Can I also assume that the worst consequence of kwin_X11 not receiving enough cpu time is frame dropping and consequently that having some X client capable to output X fps on a regular basis and kwin_X11 dropping a couple of frames from time to time would be just absurd ? ---------- Figures when running some typical desktop workload for 5 minutes: X (nice -1) ucs : 1525 / kwin ucs (nice 0) 51834 (GT=53359)
X (nice -1) ucs : 6337 / kwin ucs (nice -2) 38415 (GT=44752) ---------- ***1 :** From what I understand in xinit code, xinit (parent) is responsible for that (`setpriority(PRIO_PROCESS, serverpid, -1); as part of the startserver procedure) right after forking the child that will later exec the server...
In fact… right after… the following comment : ***don't nice server*** ??? which indeed increases even more my perplexity. ***2 :** I know that Ananicy would heavily renice (-10) kwin if not even preferring some real time scheduling model (which I do not consider sensible (*3)). There is unfortunately no rationale given for this tuning and the corresponding rule is quite old (2018), a big amount of changes have been commited to kwin since. ***3 :** At least as far as kwin_X11 is concerned since, if my understanding is correct, the X server is still managing inputs.