Dmitriy Kirhlarov wrote:>Hi Pierre! > >On Wed, 08 Jun 2005, Pierre DAVID wrote: > > > >>Do you have a clue to help us use FreeBSD and not switch on Linux >>for this service? >> >> > >Bad workaround. >You can create many small partitions and mount_unionfs. > >Actually it's a very valid choice. At this time, Linux offers ext3, XFS (from SGI), JFS (from IBM) and RieserFS as journaled file systems (as in no fscking fsck). JFS, XFS and RieserFS offer very good performance with big directories (like Maildir style mailboxes could create) and recover from unexpected outages quickly (journal replay to last checkpoint is typically seconds) and robustly. In fact, if I were to deploy a large Maildir system, where users could have thousands of files per directory, I would definitly be looking at JFS, XFS or Rieser. Linux also has a WORKING logical volume manager, and a WORKING s/w raid5 whose performance is close to all but the most high end RAID controllers. But I digress. I really like FreeBSD for a lot of things, but outside of snap and union, storage and filesystems are pretty much SOTA circa 1998 I'm not that amazed by his numbers. on a 5.3R system, when I use snap to freeze a dump of a 13% used 140 gb gmirror partition, the system (2ghz AMD, SATA150 drives) goes out to lunch for about 45 seconds, hanging all IO's to the partition. Scale that upto 1.8TB, and I could see where you could be going nowhere for a good 10 minutes just waiting for the snap to finish. Still better than waiting hours for fsck, but nowhere near the recovery speed of a true journaled system. jim
Within the space of about 10 minutes this article appeared here, On Jun 8, 2005, at 5:48 PM, secmgr wrote:> Actually it's a very valid choice. At this time, Linux offers > ext3, XFS (from SGI), JFS (from IBM) and RieserFS as journaled file > systems (as in no fscking fsck). JFS, XFS and RieserFS offer very > good performance with big directories (like Maildir style mailboxes > could create) and recover from unexpected outages quickly (journal > replay to last checkpoint is typically seconds) and robustly. In > fact, if I were to deploy a large Maildir system, where users could > have thousands of files per directory, I would definitly be looking > at JFS, XFS or Rieser. Linux also has a WORKING logical volume > manager, and a WORKING s/w raid5 whose performance is close to all > but the most high end RAID controllers. But I digress....and this one on a NetBSD list: Subject: Re: google's summer of code makefs extension To: None <netbsd-users@netbsd.org> From: Christos Zoulas <christos@tac.gw.com> List: netbsd-users@netbsd.org Date: 06/09/2005 00:18:00 In article <20050608162224.GA6831@ispid.com.pl>, Zbigniew Baniewski <zb@ispid.com.pl> wrote: >On Wed, Jun 08, 2005 at 11:01:55AM -0500, Greg Naughton wrote: >> Reiserfs would be fun. > > Will you? ;) Even that simpler (I suppose, it's simpler...) 3.5.23... Please don't. If you really want, first fix reiserfs on linux so that: 1. It has an fsck that works instead of spending 18 hours or an 1T filesystem only to proclaim in the end that it cannot be fixed. 2. Does not turn data into trash after a few weeks of heavy filesystem use. christos ----- I haven't used anything but ext2/3 on Linux, so I don't have any axes to grind one way or the other, but there definitely seems to exist some very different perspectives on how stable and robust ReiserFS is. :-) -- -Chuck PS: I'm not interested in spreading platform-specific FUD, but these two messages arriving in my inbox almost at once do make quite a contrast.
Matthias Buelow
2005-Jun-08 22:27 UTC
[Fwd: Re: Show stopper for large disks with 5.4-RELEASE]
secmgr wrote:> hanging all IO's to the partition. Scale that upto 1.8TB, and I could > see where you could be going nowhere for a good 10 minutes just waiting > for the snap to finish. Still better than waiting hours for fsck, but > nowhere near the recovery speed of a true journaled system.Seemingly Scott Long is working on journaling for ufs2. http://www.freebsd.org/news/status/report-jan-2005-mar-2005.html#Filesystem-journalling-for-UFS I like softupdates, conceptually (if they work correctly as described, which I do not know) but if ufsj can omit the fsck garbage collection after an unclean boot that would be a great boon. It would be nice if one can in the future chose between softupdates (for smaller filesystems) and journalling (for larger ones), or so. mkb.