Linux 6.12 Released With Real-Time Capabilities, Sched_Ext, More AMD RDNA4 & More

[F.Ultra]

Yes, before and after the game.

well then I have no more ideas, this should normally be either a cpu or memory related issue and it is really strange that rt_sched changes this.

[dimko]

reduced is one thing but did it show that everything else was really paused (aka taking 0% cpu)?

Yes, before and after the game.

[F.Ultra]

First thing I did - launched HTOP. Nope, CPU activity was reduced. Which is why I think issue is with polluted CPU cache. As it's only thing I can not see with usual OS tools.

reduced is one thing but did it show that everything else was really paused (aka taking 0% cpu)?

[dimko]

I agree that it is an interesting problem that makes me a bit puzzled. One idea would be to see if ctrl-z really paused all the steps in the build, aka if the compiler or make forked other processes to do various things they are not paused if the parent is paused, aka I would bring up top and see if it really is only the game taking up cpu at that point as the first step.

First thing I did - launched HTOP. Nope, CPU activity was reduced. Which is why I think issue is with polluted CPU cache. As it's only thing I can not see with usual OS tools.

[F.Ultra]

Thanks for interesting perspective.
Can't help but wonder why do I have a performance drop by using ctrl-z. Even though CPU and RAM usage did not change and CPU usage was significantly lower?

I agree that it is an interesting problem that makes me a bit puzzled. One idea would be to see if ctrl-z really paused all the steps in the build, aka if the compiler or make forked other processes to do various things they are not paused if the parent is paused, aka I would bring up top and see if it really is only the game taking up cpu at that point as the first step.

[dimko]

If you pause the execution of the compile with ctrl+z (assuming that the whole compilation was actually paused here and not just some parts of it) then there is no compilation cache, every cache line it occupied would be overwritten by the running application (aka the game). It was already in RAM so what happens is that the copy in cache is replaced.

You cannot cache data that doesn't already exist in RAM, when the CPU wants to read a peace of memory it first have to load a full cache line (64 bytes) from that position of RAM into the cache and only then can the CPU perform calculations on the data. Then when it does writes it overwrites the old position in the cache (if it updated the same data it once loaded) and then later the MMU flushes it from cache to actual RAM.

Thanks for interesting perspective.
Can't help but wonder why do I have a performance drop by using ctrl-z. Even though CPU and RAM usage did not change and CPU usage was significantly lower?

[F.Ultra]

The question becomes, where does existing compilation cache go? RAM? Something tell me - its not the case. Hence the problem.

If you pause the execution of the compile with ctrl+z (assuming that the whole compilation was actually paused here and not just some parts of it) then there is no compilation cache, every cache line it occupied would be overwritten by the running application (aka the game). It was already in RAM so what happens is that the copy in cache is replaced.

You cannot cache data that doesn't already exist in RAM, when the CPU wants to read a peace of memory it first have to load a full cache line (64 bytes) from that position of RAM into the cache and only then can the CPU perform calculations on the data. Then when it does writes it overwrites the old position in the cache (if it updated the same data it once loaded) and then later the MMU flushes it from cache to actual RAM.

[dimko]

well then it is indeed extremely strange since if the task is really paused it is paused and thus it is not running on the cpu and therefore it cannot pollute the cache at all. Who ever said that computers was easy :-)

The question becomes, where does existing compilation cache go? RAM? Something tell me - its not the case. Hence the problem.

[F.Ultra]

Can't be the case, 64 Gig of ram, and ram usage is low.

well then it is indeed extremely strange since if the task is really paused it is paused and thus it is not running on the cpu and therefore it cannot pollute the cache at all. Who ever said that computers was easy :-)

[dimko]

while the task is paused it cannot cause any cache issues so I wonder if this could simply be down to low mem situation

Can't be the case, 64 Gig of ram, and ram usage is low.