Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 09 Feb 2007 19:34:43 +0300
From:      Alexander Sabourenkov <screwdriver@lxnt.info>
To:        freebsd-stable@freebsd.org
Subject:   Re: What is a good choice of sata-ii raid controller for freebsd?
Message-ID:  <45CCA2A3.8050509@lxnt.info>
In-Reply-To: <01e601c74c5d$31be19c0$0c00a8c0@Artem>
References:  <00ad01c74b65$79db1710$0c00a8c0@Artem>	<20070208094620.GA9599@rink.nu>	<00a701c74b6e$7c3e4550$fe03a8c0@claylaptop>	<20070208165224.GA35610@icarus.home.lan>	<c7aff4ef0702080952i7abb4a4dpe15e2554019b815f@mail.gmail.com><004401c74c31$f8159160$0c00a8c0@Artem>	<45CC72D4.9040104@lxnt.info> <01e601c74c5d$31be19c0$0c00a8c0@Artem>

next in thread | previous in thread | raw e-mail | index | archive | help

> Hm... two points here. I, somehow, do not really believe that
> software raid (gmirror for example) is as reliable as hardware.
> I, deeply inside, believe that i might screw things very badly under some
> heavy load and bad timing conditions. Can't explain it. it is religious 
> i guess,
> but i can be very wrong about this.

Hardware RAID is actually a software RAID running on a processor mounted
on the card.

Just like the software RAID it can and actually did screw up badly under
some heavy load and
bad timing conditions.


> However, two perfomance point:
> Under gmirror OS must issue two commands to write to disks and some
> commands to check/set mark that mirrored data is intact.
> Under hardware RAID OS issue sonly one command to write and no
> checking command, since raid controller handles this async.
> 
> So, software OS raid must be slower than controller based raid anyway.

Hardware RAID also needs to send two commands to the disks, which they
will carry out no
faster than if they were attached to a plain SATA controller.

When disks' bandwidth is saturated, HW raid stalls just as much as SW one.

> Am i right here? Any benchmark data on this?

Benchmark data suggests that the difference is below measurement error
margin.

> As for reliability of gmirror. I just need to know how it works to see
> for myself that if power turned off in some racing condition gmirror 
> will know that
> disk are out of sync. If it is done than gmirror must check sync of 
> disks every read, and
> that mean two command for reading too, which must slow down things.
> Is it true?
> 

If you need to be absolutely sure what happens under some particular
curcuimstances,
I suggest you read the code.

For the common failure modes: hard read errors, signal cables going
loose, tripping power cords,
my experience shows that gmirror works reliably enough:  no data lost.


-- 

./lxnt





Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?45CCA2A3.8050509>