89 replies to this topic

[anudanan]

It may be a good idea to use BFQ scheduler as a new default scheduler instead of CFQ.  From my starting point with CFQ scheduler in kernel 4.1.20 I see that deadline is much more better than CFQ. The box ist now responsible and fast and as not interrupt. If BFQ is better than deadline and garantuee the realtime requirements that it will be the best choice.

But today CFQ is bad for the ARM boxes.

[anudanan]

If´ve mode today on my two uno4kse boxes with deadline.

die local box records 5 streams and the remote box records 9 streams on the same hdd of the local box. Very nice to see that it now works very well without interrupt.

 

That was not possible in my tests with CFQ.

 

I think it is exciting what the BFQ can improve in detail then.

[anudanan]

@MiLo

 

I´ve red over BFQ scheduler that it can respect realtime processes like recording threads, but these processes/threads must bei marked with a realtime label. Does the enigma package mark the realtime processes/thread correctly for using with BFQ?

[anudanan]

I´ve made today morning 20 parallel recordings from TV and a cutting job and a moving job in parallel. Everything  was ok and the box was fast enough to handle this with deadline. with cfq in kernel 4.1.20 that test was not possible to do. The recording thread became enough CPU and I/O power and the cutting/moving jobs take the rest of power

 

I´m happe now with that solution

[Dimitrij]

for D in /sys/block/sd*; do [ `cat $D/removable` == "0" ] && echo deadline > $D/queue/scheduler ; done

[anudanan]

One question for my unterstanding.

 

Why is it necessary to test `cat /sys/block/sd*/removable qual` to "0"

 

Are there other block device types which don´t have "0"?

[Dimitrij]

removable is 0 --> internal disk

[anudanan]

My external HDDs on USB1  and USB2 shows also 0 on /sys/block/sdb/removable and /sys/block/sdc/removable

 

But for me it is ok that all HDD (intern or over USB) become deadline an the I/O scheduler because all my tests with the external USB drive show the same testresults

 

cfq ist bad and deadline is fast 

[Dimitrij]

removable is 0 --> internal disk

Sorry, I was wrong.

removable is 0 --> hdd or ssd not USB flash

[anudanan]

After some weeks using the deadline I/O scheduler instead fo cfg all my problems with interrupts in recording movies during parallel HDD I/O actions like cutting or moving files are solved. The HDD performance is much more better than using cfg.

 

From my point of view it would be a good idea to switch from cfg to deadline especially for kernel 4.1.20 (uno4kse, zero4k). I can´t say if that make sense for other kernel but for me the algorithm of deadline looks for me better to support realtime requirements of recording movies than cfg.

Edited by anudanan, 8 February 2018 - 07:01.