[14 I/O Monitoring - errata] I/O scheduling with kernel 5.x?

heitorpb

The chapter 14, section about ionice ends with a warning:

ionice works only when using the CFQ I/O Scheduler, which was removed in the 5.0 kernel version.

The man page says something similar:

Linux supports I/O scheduling priorities and classes since 2.6.13 with the CFQ I/O scheduler.

But it doesn't say that ionice does not work without this scheduler.

I am running kernel 5.15.7 on ArchLinux and I can set/get different classes and priorities:

$ uname -r
5.15.7-arch1-1
 
$ ionice -p 949196 # get class and priority
none: prio 0
 
$ ionice -c 2 -n 1 -p 949196 # set class 2 and priority 1
 
$ ionice -p 949196
best-effort: prio 1

So, I don't have the CFQ scheduler, but I can use ionice. I believe this chapter needs to be update to say that it still works with kernel 5, even though the CFQ scheduler does not exist anymore.

And: was CFQ replaced with something else?

coop

what makes you think ionice is doing anything? Just because it is not returning an error doesn't mean it is having an effect. perhaps you can pass a suggestion to the ionice maintainers to have them check what scheduler is in effect and print an error otherwise. Or perhaps you can try some performance tests with varying parameters The "I/O" scheduling class has no meaning, for example, with the deadline group of schedulers.

heitorpb

what makes you think ionice is doing anything?

Hmm... So, those ionice commands I pasted above: I modified the scheduling class and I/O priority of my vim instance, but first I checked what were the values. I was able to change them. So, this means that it works, right?

Is there another way for me to double check that?

perhaps you can pass a suggestion to the ionice maintainers to have them check what scheduler is in effect and print an error otherwise.

the ionice man page does not say that it only works with CFQ scheduler, it says that it works "since 2.6.13 with the CFQ I/O scheduler." Do you have more documentation about ionice that says it doesn't work with other schedulers?

coop

https://unix.stackexchange.com/questions/136226/does-ionice-work-with-the-deadline-scheduler

https://serverfault.com/questions/485549/ionice-idle-is-ignored

https://utcc.utoronto.ca/~cks/space/blog/linux/IoniceNotes

etc with a little googling Really, if you disbelieve this just devise a test. Do a number of heavy I/O tasks and put at least one of them under ionice and do timings multiple times and compare results. With the rise of SSDs and a lot more fancy off storage, I/O scheduling really does not get fiddled with as much anymore.

In fact the course already has the lab exercise, lab_iosched.h which can easily be fiddled with to probe ionice.
My system has only [none] mq-deadline kyber bfq.

DO the experiment, it will take less time than this.

The coders of ionice just do not, for whatever reason, bother to tell you when ionice is a noop.