From mboxrd@z Thu Jan 1 00:00:00 1970 Subject: Re: [linux-lvm] JFS or XFS From: Austin Gonyou In-Reply-To: <1074209631.19419.24.camel@david.internal.NorcrossGroup.com> References: <1074209631.19419.24.camel@david.internal.NorcrossGroup.com> Content-Transfer-Encoding: 7bit Message-Id: <1074224135.2022.4.camel@localhost.localdomain> Mime-Version: 1.0 Sender: linux-lvm-admin@sistina.com Errors-To: linux-lvm-admin@sistina.com Reply-To: linux-lvm@sistina.com List-Help: List-Post: List-Subscribe: , List-Unsubscribe: , List-Archive: Date: Thu Jan 15 22:38:01 2004 List-Id: Content-Type: text/plain; charset="us-ascii" To: linux-lvm@sistina.com On Thu, 2004-01-15 at 17:33, Greg Freemyer wrote: > On Thu, 2004-01-15 at 18:22, neuron wrote: > > reiserfs, guaranteed personally. Especially on a mail server. > > I agree, reiserfs was tuned for handling lots of small files. Sounds > perfect for a mail server. Many tests prove this to be true as well. Search on google for what Greg's saying. Lots of benches out there to see. [...] > > Personally I haven't had any problems with XFS though, but I haven't > used it that much. > > > > If you go with XFS, stay away from anything older than 1.3.1. Not to mention 1.3.x is when the xfslogd and xfsdatad were finally implemented. I noted this in my prior email regarding tunable log updates. > The 1.2 and prior releases had this really nice feature that they > ignored the sync command (and fsync I think). It wasn't ignored, just not paid attention too as often it should've been. ;) > The end result is that with a power outage or kernel lockup you could > lose lots of work. I had one failure on a lightly used machine that > had > a whole days activities still sitting in the OS disk cache when the > kernel locked up. Depending on how often a system might sync()/fsync() this is true. As well as any FS options you use for mounting XFS volumes regarding log-buffers, io buffers, etc. > This is fixed in 1.3.1, > > Greg Yes, much better now. Still, even with JFS or EXT3, as I noted before, certain AbEnds will cause meta data to be lost. This is more a function of hardware caching in a non-backed state, such as large cache buffers of some drives 8MB drive cache, etc. (i.e. no battery backed up cache) -- Austin Gonyou Coremetrics, Inc.