Subject: Re: Building a huge file server
To: Andy Ruhl <email@example.com>
From: Timo Schoeler <firstname.lastname@example.org>
Date: 07/05/2006 15:14:16
thus Andy Ruhl spake:
> On 7/5/06, Daniel Carosone <email@example.com> wrote:
>> It may be heretical of me to suggest this on a NetBSD list, but for
>> something like this - and especially if you need it now - have you
>> considered opensolaris and zfs? The SATA-controller question is much
>> more difficult and fiddly there, but the filesystem questions much
>> less so. fsck on something this size takes way way way too long.
> People ask all the time if there is a journalling filesystem for
> NetBSD, and I guess this would be the appropriate time to ask if there
> is at least any progress in this area.
> All large un*x systems these days are usually not requiring fsck on a
> "dirty" reboot. If it did it would make life extremely difficult for
> some admins for sure.
> I'm sure it's way over my head to make any comments here, but in my
> own experience I've had good luck with turning off disk write cache (I
> have no idea how this affects performance) with soft updates turned
> on, and up until now I haven't had anything drastic happen. But NetBSD
> does want to do fsck anyway on reboot... I'm managing just over 1TB
> but it's all split into smaller filesystems. The biggest one is 300
> gigs. This is on older ATA100/133 drives, with NetBSD 2.0.2. No major
> problems yet, but I'm not doing anything fancy either.
i did a bit benchmarking on NetBSD-3 as well as FreeBSD 6.1-RELEASE
using a pair of 36GByte U160 SCSI drives and a pair of 250GByte SATA HDs
(WD RAID Edition) and did this for both, write cache enabled and disabled.
in both scenarios, SCSI as well as SATA, the difference was very very
small (for both, FFSv1 + softupdates and UFS2 + softupdates), so i
decided to keep write cache disabled (default configuration, at least on