All of lore.kernel.org
 help / color / mirror / Atom feed
* [Qemu-devel] Confusing I/O throttling behavior with burst_length > 1 and max < avg
@ 2016-07-14 12:18 Stefan Hajnoczi
  2016-07-26 12:38 ` Alberto Garcia
  0 siblings, 1 reply; 3+ messages in thread
From: Stefan Hajnoczi @ 2016-07-14 12:18 UTC (permalink / raw)
  To: qemu-devel; +Cc: Alberto Garcia, ngu

[-- Attachment #1: Type: text/plain, Size: 824 bytes --]

Hi Berto,
The following throttling behavior was observed with bps=512000
bps_max=51200 throttling.bps-total-max-length=2:

  READ: io=512KB, aggrb=50KB/s, minb=50KB/s, maxb=50KB/s, mint=10061msec, maxt=10061msec

If bps-total-max-length=1 then the result is different:

  READ: io=5056KB, aggrb=505KB/s, minb=505KB/s, maxb=505KB/s, mint=10006msec, maxt=10006msec

The burst limits guest bps instead of adding on top of it when
burst_length > 1.

I'm not sure what behavior you intended?  Replacing the following should
allow the guest to get its burst on top of the bps:

  return throttle_do_compute_wait(bkt->max, extra);

with:

  return throttle_do_compute_wait(MAX(bkt->avg, bkt->max), extra);

What do you think?

For more details, see Nini Gu's bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=1355665

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [Qemu-devel] Confusing I/O throttling behavior with burst_length > 1 and max < avg
  2016-07-14 12:18 [Qemu-devel] Confusing I/O throttling behavior with burst_length > 1 and max < avg Stefan Hajnoczi
@ 2016-07-26 12:38 ` Alberto Garcia
  0 siblings, 0 replies; 3+ messages in thread
From: Alberto Garcia @ 2016-07-26 12:38 UTC (permalink / raw)
  To: Stefan Hajnoczi; +Cc: qemu-devel, ngu

On Thu, Jul 14, 2016 at 01:18:19PM +0100, Stefan Hajnoczi wrote:

> The following throttling behavior was observed with bps=512000
> bps_max=51200 throttling.bps-total-max-length=2:
> 
>   READ: io=512KB, aggrb=50KB/s, minb=50KB/s, maxb=50KB/s, mint=10061msec, maxt=10061msec
> 
> If bps-total-max-length=1 then the result is different:
> 
>   READ: io=5056KB, aggrb=505KB/s, minb=505KB/s, maxb=505KB/s, mint=10006msec, maxt=10006msec
> 
> The burst limits guest bps instead of adding on top of it when
> burst_length > 1.

Thanks Stefan,

I think we should forbid that kind of setup altogether. I wrote a few
more details in the bug report:

> https://bugzilla.redhat.com/show_bug.cgi?id=1355665

I can write the patch.

Berto

^ permalink raw reply	[flat|nested] 3+ messages in thread

* [Qemu-devel] Confusing I/O throttling behavior with burst_length > 1 and max < avg
@ 2016-07-14 12:18 Stefan Hajnoczi
  0 siblings, 0 replies; 3+ messages in thread
From: Stefan Hajnoczi @ 2016-07-14 12:18 UTC (permalink / raw)
  To: qemu-devel; +Cc: Alberto Garcia, ngu

[-- Attachment #1: Type: text/plain, Size: 824 bytes --]

Hi Berto,
The following throttling behavior was observed with bps=512000
bps_max=51200 throttling.bps-total-max-length=2:

  READ: io=512KB, aggrb=50KB/s, minb=50KB/s, maxb=50KB/s, mint=10061msec, maxt=10061msec

If bps-total-max-length=1 then the result is different:

  READ: io=5056KB, aggrb=505KB/s, minb=505KB/s, maxb=505KB/s, mint=10006msec, maxt=10006msec

The burst limits guest bps instead of adding on top of it when
burst_length > 1.

I'm not sure what behavior you intended?  Replacing the following should
allow the guest to get its burst on top of the bps:

  return throttle_do_compute_wait(bkt->max, extra);

with:

  return throttle_do_compute_wait(MAX(bkt->avg, bkt->max), extra);

What do you think?

For more details, see Nini Gu's bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=1355665

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 473 bytes --]

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2016-07-26 12:39 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2016-07-14 12:18 [Qemu-devel] Confusing I/O throttling behavior with burst_length > 1 and max < avg Stefan Hajnoczi
2016-07-26 12:38 ` Alberto Garcia
  -- strict thread matches above, loose matches on Subject: below --
2016-07-14 12:18 Stefan Hajnoczi

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.