Re: RELENG_4 -> 5 -> 6: significant performance regression

From: Kris Kennaway <kris_at_obsecurity.org>
Date: Thu, 27 Apr 2006 21:01:46 -0400
On Thu, Apr 27, 2006 at 08:42:47PM -0400, Brian K. White wrote:
> 
> ----- Original Message ----- 
> From: "Bartosz Fabianowski" <freebsd_at_chillt.de>
> To: <ml_at_sd2i.com>
> Cc: <freebsd-stable_at_freebsd.org>
> Sent: Thursday, April 27, 2006 3:40 PM
> Subject: Re: RELENG_4 -> 5 -> 6: significant performance regression
> 
> 
> >>You wrote that Giant is needed in 6.0 and now you write it has been
> >>removed.
> >
> >In 4.x, every UFS write requires the Giant lock. In 6.x, Giant is not 
> >normally required, making file system operations faster. When you enable 
> >QUOTA, you basically get back to the 4.x behavior where Giant is needed 
> >for each write. This is why in 6.x UFS will normally be faster but if you 
> >enable QUOTA, you lose the newly gained performance again.
> 
> In that case the test was correct the question stands at least on that 
> point.
> If quota isn't mpsafe yet, that's fine. It just means that performance 
> should still be only about the same as on 4 while quota is enabled.
> Removing a feature you had before is no answer to this question.
> Removing an expensive new feature that didn't exist before could be.

The above isn't quite accurate because 4.x uses the spl model which
means that interrupts may be serviced while other code is operating at
a lower spl level.

The OP still needs to test without quotas (and other changes
mentioned) to see whether it has an impact on his benchmark.

Kris

Received on Thu Apr 27 2006 - 23:01:48 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:55 UTC