vinum, 5.x performance, ... (was: Re: ANNOUNCE: GNOME 2.12.1 has been merged into the ports tree)

[Søren Schmidt]

On 08/11/2005, at 10:16, Alexander Leidinger wrote:

> Ion-Mihai Tetcu <itetcu at people.tecnik93.com> wrote:
>
>> On Mon, 07 Nov 2005 14:56:45 +0100
>> Alexander Leidinger <Alexander at Leidinger.net> wrote:
>>
>>> Ion-Mihai Tetcu <itetcu at people.tecnik93.com> wrote:
>>>
>>> > On Mon, 07 Nov 2005 11:18:00 +0100
>>> > Alexander Leidinger <Alexander at Leidinger.net> wrote:
>>> >
>>> >> > - and performaces is a nightmare on atapi harddisks
>>> >>
>>> >> AFAIR 4.x enables the write cache (bad if there's a power failure
>>> >> in the soft-updates case, since without native command  
>>> queueing the
>>> >> drivers may reorder writes witout the possibility to notify
>>> >> soft-updates about it),
>
> I' a little bit confused now. The man page says the WC is enabled  
> by default,
> but I was under the impression, that it is disabled by default (I  
> remember
> there was a discussion about this long ago). On my systems it is  
> disabled,
> since I haven't removed the loader.conf line since the 4.x days.  
> Søren, is
> this a bug in the man page or did I got it wrong?

WC is enabled by default on modern FreeBSD's, it was tried to disable  
it once but the performance hit quickly changed that decision.
>>> > So on NCQ disks write_caching should be safe from consistency
>>> > point-of-view ?
>>>
>>> AFAIK: Enabling the write cache doesn't mean you use NCQ. So just
>>> because your drive is able to do NCQ, you shouldn't enable the write
>>> cache. If your controller, your disk and the OS is able to use NCQ,
>>> and you enable the use of NCQ, then everything is fine... assuming
>>> the drive doesn't lie to you. AFAIR there are or there where some
>>> drives which lie(d), but I don't remember the details.
>>
>> And finding out the what disk and controller use NCQ _and_ are
>> supported by our drivers can be done how ?
>
> Asking sos at FreeBSD.org (CCed) would be a way to find it out...

NCQ is not supported on FreeBSD at this time, support is in the works  
and will get into -current once its far enough along. Note that TCQ  
is only supported (and buggy since the PAE code was added) in 4.x.

> At first I thought atacontrol tells you about it. But the  
> "atacontrol cap
> ad0" command just tells you about the capabilities of the drive,  
> not which
> of those capabilities are used by the driver (would be nice to have,
> specially in the light of seeing TCQ enabled but knowing that you  
> said is
> doesn't work... at least with the previous implementation of ata).
>
> Søren, is there a way to en-/disable some of the capabilities with  
> atacontrol
> or something else?

Not as is, but it could be added along the way. However some of the  
capabilties depends on both disk and controller and in most cases  
also the OS have support for it. Furthermore some if the capabilities  
only makes sense to change during boot.

Søren Schmidt
sos at FreeBSD.org