Date: Thu, 20 Sep 2001 23:15:28 -0500 From: Mike Meyer <mwm@mired.org> To: "Daniel O'Connor" <doconnor@gsoft.com.au>, "Kevin Oberman" <oberman@es.net> Cc: stable@freebsd.org Subject: RE: hw.ata.wc && hw.ata.tags && softupdates short question Message-ID: <15274.48864.994229.51687@guru.mired.org> In-Reply-To: <XFMail.20010921121832.doconnor@gsoft.com.au> References: <XFMail.20010921121832.doconnor@gsoft.com.au> <200109210329.f8L3TGR30860@ptavv.es.net> <20010920211704.C7820@gateway.bogus>
next in thread | previous in thread | raw e-mail | index | archive | help
Daniel O'Connor <doconnor@gsoft.com.au> types:
> On 20-Sep-2001 Nuno Teixeira wrote:
> > For what I heard, I concluded that we shouldn't use softupdates with write
> > cache turned on. The first time that I tried this I loose a lot of work
> > due to a power failure.
> You shouldn't use write caching, period.
Matt Dillon disagrees. From the tuning(7) man page:
There is a new experimental feature for IDE hard drives called
hw.ata.tags (you also set this in the bootloader) which allows write
caching to be safely turned on. This brings SCSI tagging features to
IDE drives.
In other words, write caching is safe if the drive and driver supports
tagged queuing. For IDE drives, Matt goes on to say:
As of this writing only IBM DPTA and DTLA drives support the
feature. Warning! These drives apparently have quality control
problems and I do not recommend purchasing them at this time. If
you need performance, go with SCSI.
> No, write caching tells the drive "Please lie to me about command completion"
> so when the OS performs a write the drive caches it and says "yes that's on
> disk", when it isn't.
Tagged queuing allows the writes to complete in a different order than
they were issued. This allows the drives to both have an effective
write cache, *and* to notify the controller when sectors are actually
on the disk.
Kevin Oberman <oberman@es.net> types:
> 1. IBM drives that do tagging are claimed to be safe. The way tagging
> works should assure that the critical metadata is always written to
> disk.
To clarify, it isn't the drive that knows about critical
metadata. That's softupdates doing it's job. Softupdates has to trust
the drive to make sure the critical metadata is on the disk. This is
why write caching sans tagged queuing and softupdates are such a bad
idea. This is also why a file system with soft updates enabled simply
ignore the async flag on any mount request.
> It's also worth noting that Soren reversed himself and made wc default
> a few months after 4.3 was released. I assume it defaults to "on" in
> 4.4, although I have not checked. This makes me suspect that he
> decided that the risk was reasonable. But I really should not speak
> for him.
If you wade through the mail lists during the 4.3 beta and release
periods, what you'll find is that disk performance took a major hit -
some people claimed as much as a factor of 6. No one discussed any
change in the risk assessment. I think he turned it back on for the
same reason he turned it off: "popular demand" <URL:
ttp://www.FreeBSD.org/cgi/getmsg.cgi?fetch=51636+0+/usr/local/www/db/text/2001/freebsd-mobile/20010318.freebsd-mobile
>.
<mike
--
Mike Meyer <mwm@mired.org> http://www.mired.org/home/mwm/
Q: How do you make the gods laugh? A: Tell them your plans.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?15274.48864.994229.51687>