Lethargic rpi3, but seemingly still running

[Jamie Landeg-Jones]

bob prohaska <fbsd at www.zefox.net> wrote:

> Sorry, I was referring the %idle number above the bargraph and inverting the 
> fraction, without noticing the %busy in the Disks output.

Ah, ok! Yes, it was the I/O I was concerned with, not the CPU.

> At the moment the same machine is compiling clang8 with about 900M of swap in use,
> here is another sample:
> Disks mmcsd   da0 pass0                               264 intrn       cpu3:ast
> KB/t   5.51  5.77  0.00                            169680 wire    641 cpu0:pree
> tps     461   457     0                            645096 act    1612 cpu1:pree
> MB/s   2.48  2.58  0.00                              2280 inact  1522 cpu2:pree
> %busy    77    50     0 
>
> Swap is on both mmcsd and da0 (Samsung Evo+ microsd and a Sandisk usb3 flash drive). 
>
> > However, I suspected something like that - high %busy whilst little actual
> > transfers taking place. I've seen it before, along with the characteristics
> > you describe, but unfortunately I don't have a fix.
> > 
>
> So far I think the problem has only been obvious when bsdtar is running. If
> there's another way to provoke it that might be interesting to try.

Unfortunately, my Pii's are not working at the moment (one sdcard broke, the other
sdcard was "borrowed" for something else!) so I can't test this, but my
suspicion is that it's related to many file creations (and/or deletions?)

You could try a script that creates lots of small files.. Do you have soft-updates
enabled? And do you have 'trim' enabled for the sdcard?

I didn't mention this in earlier replies, because I cannot find any reference to
back it up, but here goes, take it with a pinch of salt!:

>From what I understand, ARM I/O causes blocking of the CPU. I know on my
android 'desktop' devices that if I have heavy I/O, everything else (even the mouse
cursor) can freeze for 10s of seconds at a time, whilst the CPU is otherwise idle.

At a guess, this problem gets worse with a fragmented or aging sdcard/usb stick.

Does any of this match your observations?

> Thanks for reading more closely than I wrote!

:-)

Cheers, Jamie